2026년 7월 6일

0.10.17Latest release2026년 7월 6일

Monterey Launch, Settings, and Subscription Sign-In Fixes

This hotfix restores launch on supported older Macs, steadies Settings plan loading, and fixes a reconnect loop for ChatGPT Subscription and GitHub Copilot providers.

Highlights

Mem now launches correctly on macOS 12 Monterey and macOS 13.
ChatGPT Subscription reconnect now starts a real fresh sign-in instead of showing connected while background AI asks you to reconnect.

Fixed

The in-app startup check now matches Mem's supported macOS range. Users on macOS 12 Monterey and macOS 13 no longer get stopped by an incorrect “system not supported” screen.
Settings no longer gets stuck endlessly checking your plan. The plan panel now loads once and settles instead of re-requesting license status every render.
ChatGPT Subscription and GitHub Copilot sign-in state is now reset when you reconnect from Settings. If an old session has expired, Mem stops treating it as connected and starts a new device-login flow when you click Authenticate.
Provider status in Settings is stricter for subscription providers. A saved model alone no longer shows ChatGPT Subscription or GitHub Copilot as ready unless the account is actually signed in.

0.10.16

2026년 7월 6일

0.10.162026년 7월 6일

Safer Background Catch-Up, Remote Recovery, and Large Uploads

This hotfix makes large Knowledge Graph catch-up work pace itself correctly on self-hosted and long-running libraries, keeps desktop clients connected to their remote Mem after settings drift during an update, and fixes large backup uploads through the web/remote app.

Highlights

Large Knowledge Graph backlogs now pause between bounded batches instead of continuously calling your model.
A desktop client connected to a self-hosted Mem is less likely to fall back to a fresh local server after an update.

Fixed

Background Knowledge Graph catch-up now treats a full 50-memory batch as a signal that more backlog may remain. Mem schedules one delayed follow-up and respects that wait from every background entry point before making another model call.
Failed Knowledge Graph batches now stay on the normal failure backoff path. Mem no longer retries a failed full batch faster just because it filled the scan window.
App logs now show the actual background task behind each scheduler model call. The log line no longer labels unrelated background AI work as KG extraction, which makes model-usage reports much easier to diagnose.
ChatGPT Subscription and GitHub Copilot sign-in now stay in Settings. If a subscription session expires, background AI asks you to reconnect the provider instead of opening unattended device-code prompts or retrying provider 429 errors in a tight loop.
Background AI defaults are quieter and the Processing settings now match what Mem actually runs. Low-signal reviews default to weekly checks, new installs use a 48-step default, and the step limit tops out at the real 64-step execution ceiling instead of showing a 100-step value that would be silently clamped.
Background Work now drains older synced-conversation review queues left by previous scheduler versions. If the panel shows many synced conversations waiting, Mem turns them into real bounded review work, clears entries that were already handled, and checks the next wave every minute instead of leaving the count frozen.
Memory reads now return the saved importance score consistently. After you update a memory's importance, `nmem memories show --json` and `nmem memories list --json` report the real value instead of `0.0`.
Large ChatGPT, Alma, and ChatWise backup uploads now use the configured server upload limit in the web and remote app too. Self-hosted imports no longer hit a hidden 2 MiB gateway limit and return HTTP 413 from the Mem app log after the Docker limit has already been raised.
Imported conversation details now show the messages that Mem already saved. If a conversation appeared in the list and could be searched or deleted, opening it, exporting it, or reading it from MCP/CLI now shows the same transcript instead of an empty message list.
OpenCode conversation sync now handles OpenCode's numeric SQLite and JSON timestamps correctly. `nmem t sync --from opencode --apply` no longer fails with a timestamp 422 before creating the thread.
`nmem plugins check` now recognizes the current Claude Code marketplace plugin layout and reports the installed version for `nowledge-mem@nowledge-community`. The older skills-folder install remains supported as a fallback.
AI Now code blocks now have a copy button, including code shown inside richer markdown responses. Generated scripts and commands are easier to reuse without selecting the block by hand.
Skill update and strengthening prompts now show in-card progress after you click them. The button changes state while Mem applies or rewrites the skill, instead of relying only on a toast.
Full search reindex can now repair sources left in `extracted` after an interrupted ingestion, as long as their durable chunks already exist and were rebuilt into the search index. Half-written sources without chunks are still left untouched.
Desktop remote clients now recover from their existing CLI remote configuration if the app's own connection settings are missing or partially rewritten during an update. Mem uses that as a read-only fallback, so a Mac or Linux client keeps pointing at its self-hosted server instead of unexpectedly starting a fresh local backend or quick tunnel.
Desktop startup now only skips the local backend when the saved remote connection is complete. If the app settings keep only the server URL but lose the key, Mem falls back to the CLI recovery path or starts locally instead of opening half-connected.
App startup is more tolerant of partial settings files. Missing shortcut or version fields no longer invalidate the saved remote connection.
Access Anywhere is safer across app restarts. Mem now prevents its managed tunnel from keeping the local database locked after the backend exits, and cleans stale local backend listeners before starting a replacement.
OAuth sign-in checks now fail cleanly when a provider does not support device-flow sign-in. Supported GitHub Copilot and Codex sign-ins are unchanged, while unsupported provider probes no longer leave a background sign-in task waiting.
Timeline's Show more button now loads the next page of activity directly instead of re-reading a larger copy of the first page. Older activity appears more reliably, even when retired feed rows or grouped background events sit between the current view and the next visible item.
Stats now makes Memory Health catch-up easy to start. When Mem has relationship reviews to catch up on, the Memory Health card shows a clear Catch up today action, gentler 3-day and 7-day options, and live remaining-count progress while it runs.
Timeline conflict reviews now show the full two-memory comparison in the right panel. Mem loads the complete memory text when available, labels the existing and new memories, and adds a concise diff preview so you can decide without guessing from truncated snippets.
Timeline conflict follow-ups now include the selected review card and the two related memories. When you tell Mem what to do, the agent sees the exact conflict and can use the same safe resolution path as the visible review buttons instead of answering from a generic Timeline chat.
Every visible Timeline review choice now works from the agent follow-up too. If you tell Mem to keep the older memory as latest, it uses the same safe resolution path as the card button instead of rejecting the instruction.
Advanced Knowledge Graph catch-up overrides are more predictable. Setting the catch-up cooldown to zero now means "continue immediately" instead of accidentally turning off the follow-up check.
Daily briefing items are easier to read from Timeline. Discoveries, crystals, and attention items now open a full right-panel detail, so important text is no longer hidden behind a short preview.
AI Now reasoning cards feel steadier when opened or focused. Mouse clicks no longer leave a bright system outline, keyboard focus uses a subtle in-card ring, and expanded reasoning keeps consistent top spacing while it folds and unfolds.
Connected Skills are more reliable on Windows and in tools such as Hermes Agent. Mem now uses real copied skill folders on Windows, verifies that each connected skill's `SKILL.md` is readable before marking it live, and replaces old unreadable Mem-created links on the next sync.
Settings → Models now shows local model storage more accurately. The Local LLM card reports the full Gemma local package instead of only the vision projector, and the Clear Cache size uses the same model files the clear action will actually remove.

0.10.15

2026년 7월 5일

0.10.152026년 7월 5일

Self-Hosted Stability, Space Recovery, and Cleaner AI Now

This patch keeps self-hosted and headless servers stable under heavy background work, brings back spaces and memories that could disappear after an update, and makes AI Now easier to follow during longer agent runs with cleaner plugin settings and history cleanup.

Improved

Selecting many Library files is much faster. Shift-click a file's checkbox to select a whole range, or use Select all and Clear in the list footer, then run a batch action on the selection (start an AI Now session, delete, or re-extract). Clicking files one at a time still works.
Deleting an AI Now task is less disruptive. The confirmation now opens inside the task row instead of dimming the whole workspace, with shorter copy and a clear preview of the local folder that will be removed.

Fixed

AI Now no longer shows the same plan twice after an approval step. When the agent has both a structured plan update and a plan tool card, Mem keeps the richer progress card and hides the duplicate checklist.
The active AI Now plan stays visible while the agent starts its next tool call, so progress is easier to track when Memory Search or another tool appears below it.
AI Now history repairs itself if the local session index is damaged. Recent tasks whose messages are still on disk appear in History again, and creating or naming a task no longer fails with a "No such file or directory" error when several session updates happen at once.
Older AI Now tasks can be removed from History again. Tasks migrated from older builds, or recovered from message files after a catalog repair, no longer fail with a "session not found" error when you delete them.
AI Now plugin settings now apply to the active task immediately. Saving an Obsidian vault path, API key, or OAuth state updates the current task and refreshes plugin context on the next message.
Memories no longer show `unknown` as a source when older records only have missing or placeholder source metadata. Memory cards and details now hide absent source data instead of presenting it as a real origin.
Stats → Memory Sources and the Memories source filter now follow the same source rules as memory cards. Distilled memories count under their original app, older MCP rows use the calling app when available, and missing placeholder sources stay out of the chart.
Turning skills off for a connected tool works for every tool again. Disconnecting skills for some tools (such as Proma) failed with an "unknown host" error because the tool was missing from an internal list. The supported-tool list now comes from one source, so connecting and disconnecting skills works across all supported tools.
Self-hosted and headless servers stay stable under heavy background work. On smaller or memory-limited machines, background knowledge-graph tasks (community detection, PageRank, knowledge extraction) could run the graph database out of memory, which showed up as server errors or a disconnected database and caused extra AI usage from repeated retries. Mem now sizes its graph memory to the machine (including container limits), eases off instead of retrying in a loop when memory is tight, and skips a single heavy pass for one cycle rather than taking the whole database down. Background intelligence keeps running and catches up on its own.
Your spaces and their memories come back after an update. On some upgrades the space list could reset to just Default, which hid the memories filed under your other spaces (they were never deleted). Mem now protects the space list from being overwritten, restores any spaces that still hold memories when the server starts, and keeps a backup of your settings before each change.
Saving preferences is more reliable. Settings changes now save through the backend as intended, so a busy console no longer fills with request errors when you open Settings.

0.10.14

2026년 7월 4일

0.10.142026년 7월 4일

Fresher Memory Context, Steadier Background Work, and Safer MCP Connections

This patch tightens the 0.10 native engine around the places users felt most: Working Memory refreshes reliably again, background intelligence respects the off switch and provider cooldowns, Timeline keeps itself fresh on Linux and remote sessions, custom provider URLs stay under your control, and MCP clients connect with clearer identity and fewer setup traps.

Highlights

Background Intelligence now stops when you turn it off, and backs off clearly when your provider is rate-limited.
Working Memory refreshes again even when Mem decides not to publish a daily briefing card.
Timeline stays fresh on Linux, remote, and suspended-webview sessions even if the live stream drops.
Custom provider Base URLs are preserved exactly, so Mem no longer silently appends `/v1` to gateways that do not use it.
Skills and MCP traffic now show what actually happened, so a reviewed skill or a connected MCP client no longer looks stuck or silent.

Fixed

Working Memory stays fresh again. A scheduled daily briefing that decides there is no grounded briefing to show no longer marks Working Memory as freshly updated. The rolling refresh can keep running, so session-start context no longer gets stuck on old dates.
Memory lists now include creation time in the same shape as memory details. CLI and MCP clients can again tell which memories are recent instead of seeing blank timestamps in list responses.
Mem-hosted MCP connectors are easier to approve. Remote MCP OAuth now understands the `/mcp` discovery paths some clients probe, and AI Now opens Mem's browser-pairing step before asking you to approve a Mem MCP connector.
Turning off Background Intelligence now fully stops low-priority knowledge-graph backfill work. A graph backfill can no longer keep calling your model after the main switch is off.
Provider rate-limit waits now use the real cooldown and are shown as provider waits in Background Work. When a service such as Kimi is over quota, Mem pauses and resumes later instead of making the queue look like a normal hourly batch wait.
Background knowledge-graph work now paces itself. When there is a large backlog to catch up on, Mem works through it in short bursts spaced over time instead of running flat-out, so the app stays fast while you use it and background work does not use far more of your AI than it should. A big backlog still finishes on its own, just gradually and out of the way.
Timeline keeps a gentle refresh path behind the live event stream, so Linux, remote, and suspended-webview sessions no longer need a manual refresh just to see recent saves or background events.
Timeline review choices are steadier while you scroll. The inline “Tell Mem what to do” box keeps Timeline scrolling under the mouse wheel, and review-action safety notes are shown in the card instead of disappearing with hover-only tips.
Memory Health catch-up buttons now save the catch-up plan even if the immediate queue kick cannot run, and the error message tells you when background work needs a restart.
Custom OpenAI-compatible provider Base URLs are no longer rewritten at save time. Known providers such as Ollama and Lemonade still get the right default, but custom gateways keep the exact API root you enter.
Search index rebuilds tolerate NUL bytes and unsafe control characters before text reaches a local llama.cpp embedder, so one bad memory or imported source no longer stops the whole rebuild.
Skill cards clear their attention marker after you open and review the activity. Mem now separates unread activity from long-term skill quality, and active skills show a real usage count from connected tools when one exists.
The Skills overview now separates skills that are already built and ready to turn on from new suggestions that are still waiting for review. Counts now match the way skills appear in the local skill folders instead of making inactive skills look over-counted.
MCP live traffic now shows client discovery as well as tool calls. Clients such as Hermes Studio that connect and read the tool list no longer leave the traffic panel empty before the first tool call.
Spaces stay visible across the app and CLI after settings changes. Non-default spaces no longer disappear from `nmem spaces` or show as empty just because the app saved the roster in its newer shape.
Your knowledge graph fills up again on cloud models. Running on a cloud or managed model with no local model loaded used to extract nothing, leaving the graph, entity search, and relationships empty. Those entities are now kept, and the extraction reads more like a knowledge graph than a flat list of names: it keeps the systems, decisions, and conditions worth reasoning over.
Saving a conversation no longer piles up duplicate copies of your project's standing instructions. When a coding tool injects a repo's AGENTS.md or CLAUDE.md into the chat, that setup text is treated as context instead of being distilled into a new memory every session.

Improved

arXiv `abs`, `pdf`, and `e-print` links now prefer arXiv's structured HTML version when it is available. Papers keep cleaner sections, equations, and figure context than PDF text extraction, while the original arXiv URL remains the source.
The hourly Background Work limit is now described as a quota, cost, and device-heat guard, not a quality setting. Raising it may make Mem catch up faster, but it can spend more provider quota.
The Background AI usage overview now keeps monthly usage and the rolling 24-hour limit clearly separated. Month totals are rebuilt from the retained daily ledger when possible, so the month card no longer collapses into the 24-hour number.

Platform

Linux desktop startup now handles WebKit rendering more carefully on Wayland. NVIDIA Wayland sessions keep DMA-BUF available and use a safer sync workaround, while user-set desktop environment variables still take priority.
The animated app icon in Settings no longer shows native video controls over the About page on Linux/WebKit builds.

0.10.12

2026년 7월 3일

0.10.122026년 7월 3일

Spaces That Delete Reliably, a Steadier Background, and Clearer Memory Origins

This patch makes removing a space actually stick, stops a background analysis from crashing the app on larger graphs, restores the original source and icon on distilled memories, and also makes command-line restores and AI Now's local-provider workflow more reliable.

Improved

AI Now feels steadier during long tool-heavy tasks. The transcript no longer fights you when you scroll up, tool output causes less repaint work, approval requests sit directly above the composer instead of taking over the whole screen, web search results no longer appear as Code cards, and typing “continue” after an interrupted task now gives the agent explicit recovery context.

Fixed

Deleting a space now sticks. Removing a space could previously reappear after a refresh, because the app and the backend both wrote the workspace list and could race each other. The backend is now the single owner of that list, so a removed space stays removed and your other spaces are never overwritten by an out-of-date save.
On larger graphs, a scheduled background analysis could run the graph engine out of memory and crash the backend into a restart loop, which in rare cases dropped spaces from your workspace. Mem now sizes that working memory to your graph, and when memory is tight it safely skips the analysis instead of crashing, so background work can no longer destabilize your data.
Memories distilled from a connected tool show that tool's icon and source again. A recent change surfaced only the Background Intelligence label and dropped the origin, so a memory distilled from a tool like Hermes now shows both where it came from and that Background Intelligence created it.
`nmem export` / `nmem import` restores no longer fail with a Kuzu STRING-to-BOOL parameter error when imported graph rows contain fields in different orders. Mem now binds each imported property by its field name instead of its position in the row, so a large restore can finish instead of stopping partway through.
AI Now approval prompts close cleanly when a request has already been handled or the turn has moved on. Repeated clicks no longer leave the permission dialog stuck on screen.
Timeline additions that fail during local-provider processing now remain visible as a clear timeline error instead of disappearing after the temporary Thinking row is removed.
AI Now stays lighter while a local-provider task is running. Background maintenance now waits for the interactive turn to finish, streaming text updates repaint less often on Linux WebKit, and high-core desktop machines no longer start the backend with an unnecessarily wide async runtime by default.
Stats Overview no longer leaves Memory Health as endless loading placeholders if the maintenance status request takes too long. The card now shows a retry state while the rest of Stats stays usable.
The local graph visualization page used by the API docs works again on the Rust engine. Opening `/graph/vis` now serves the bundled graph explorer instead of returning 404.
Access Anywhere now preserves auto-reconnect across app quit, backend shutdown, and app updates. Mem closes the managed Cloudflare helper cleanly without treating that lifecycle event as you turning Access Anywhere off.
Background AI no longer keeps hitting a rate-limited model. When a background task runs into a provider rate limit or quota (for example a Kimi Code subscription's hourly cap), Mem pauses background work and backs off instead of retrying every couple of seconds, so it stops burning your subscription allowance. The pause grows on repeated limits, respects the provider's retry-after hint, and recovers on its own once the window clears.
Background AI now works with reasoning models that reject a temperature setting, including GPT-5 class models and setups pointed at ChatGPT's Codex endpoint. When the provider rejects the temperature, Mem retries the same request without it automatically, so knowledge extraction and distillation no longer fail against those models.

Added

You can now correct a memory's origin in place. Editing a memory accepts a source value, so a mis-detected origin (for example a memory wrongly tagged as agent-created) can be fixed without deleting and recreating it, which used to reset the memory's creation date, history, and relationships.
AI Now now includes Lemonade as a local MCP integration. If you run Lemonade Server locally, you can enable it from Plugins instead of adding the MCP endpoint by hand.
Custom providers can now choose the Anthropic Messages protocol alongside Chat Completions and Responses API, making Anthropic-compatible self-hosted or proxy endpoints easier to configure correctly.

0.10.11

2026년 7월 3일

0.10.112026년 7월 3일

Clear Access Settings, Safer Restores, and Smoother Remote Use

This patch makes command-line server access settings easier to trust, makes export/import restores more resilient when graph relationship fields overlap with internal import parameters, helps large Data Transfer exports finish without unnecessary temporary disk pressure, reduces repeated AI Now setup work when reopening the tab, and makes headless remote setup easier from the web settings page.

Fixed

`nmem config access show` now matches its JSON output. If Same-Wi-Fi access is enabled, the readable view says it is on instead of incorrectly showing `off (127.0.0.1)`.
`nmem export` / `nmem import` restores no longer stop partway through when graph relationship properties share a name with import parameters. Mem now keeps those parameters separate so Kuzu does not confuse strings and booleans.
Data Transfer exports handle large thread histories more gracefully. Mem now streams the message store into the export instead of holding the whole table first, and storage-space errors show as a readable message instead of raw JSON.
AI Now does less repeated setup work when you leave and reopen the tab. Skill prompt sync and plugin-version checks now run only when they are actually needed, which reduces avoidable stalls on slower Linux WebKitGTK/Wayland render paths.
Access Anywhere settings can now copy the current server key from a remote or web settings page when that key is available to the client. This makes headless servers easier to finish setting up without switching back to SSH just to find the key.
The Memory Health card is calmer. A relationship-review backlog that Mem is comfortably keeping up with no longer reads as high maintenance pressure; the card only flags pressure when it is genuinely falling behind.

Added

Catch up on background review on your terms. When memories are waiting for relationship review, the Memory Health card lets you pick how fast to clear the backlog: Today (all at once), 3 days, or 7 days. It shows live progress with a Stop button and clears it in bounded background waves, so a large backlog no longer means a scary multi-week wait.

0.10.10

2026년 7월 2일

0.10.102026년 7월 2일

New Spaces Stay Put, Reliable Linux Installs, and Clearer Timeline

This patch fixes the rough edges found after 0.10.9. A Space you just created no longer disappears, Linux .deb and .rpm installs start reliably, Deep Research reads more pages, background flags and insights speak your language, Timeline failures stay visible, Lemonade is a first-class local provider, and quitting or updating checkpoints the graph more reliably.

Fixed

A Space you just created no longer disappears. A background task that reviews your guidance rules could save an older snapshot of your settings and quietly drop a Space made moments earlier. Your memories and threads were always safe, and now the Space stays too.
Linux .deb and .rpm installs start reliably. On some Debian, Ubuntu, and Fedora systems the app reported that it could not find its backend and never opened; it now looks in the right place for packaged installs.
Deep Research gathers more sources. Fetching a page could fail with a one-time connection error on sites that reset a cold connection; Mem now retries briefly before giving up, so research reads the page instead of falling back to model knowledge.
Automatic flags, insights, and crystals speak your language. Their titles and summaries now follow your preferred language instead of defaulting to English.
Asking Mem about a Timeline flag streams its answer as it thinks, and you can keep asking follow-up questions on the same card instead of it resetting after one reply.
Timeline agent failures no longer vanish. If a provider or tool call fails while adding something to Timeline, Mem leaves a visible entry with the failure message instead of removing the temporary row after the stream ends.
Lemonade is now a first-class local provider. Choose Lemonade in Providers, use its default local `/v1` endpoint, and fetch models without configuring it as a generic custom endpoint first.
AI Now can use custom HTTP MCP tools again. Enabled custom MCP servers are connected to the agent with their configured headers or API key auth, so local tool servers such as Lemonade can be added from the Plugins panel.
Background AI works with stricter OpenAI-compatible chat providers. AI Now and background intelligence now send system instructions in the safest Chat Completions shape, so providers that reject `input_text` no longer fail the whole task.
Quitting or updating while Mem is busy is safer. The server now checkpoints through both the old checkpoint URL and the shutdown path, reducing the chance that recent graph writes are left waiting in the WAL.
Skills imports handle real files better. CJK descriptions, quoted YAML text, common words like `task-brief`, and frontmatter-only name or description edits now work instead of being rejected or treated as unchanged.
Skill detail pages are less likely to open empty. If a skill's main file is missing after an upgrade or import but the installed copy still exists for your AI tools, Mem now shows the readable skill text instead of a blank details area.
Search rebuilds are more resilient to corrupted imports. NUL bytes in saved sources, chunks, memories, entities, communities, or thread messages are cleaned before indexing, so one bad record no longer blocks the rebuild path.
Knowledge Graph extraction no longer hides parse failures as empty success. If the model returns something that is not valid KG JSON, Mem leaves it as a failed attempt instead of marking the memory extracted-empty.
Space and plugin UI warnings are quieter and more accurate. Changing only a Space icon keeps the display name, disabled OAuth plugins no longer count as needing authentication, and `nmem models status` matches the JSON search-embedding status.

0.10.9

2026년 7월 2일

0.10.92026년 7월 2일

Big Imports That Finish, and AI That Recovers

Bringing in a large history no longer crashes the app partway through, and self-hosted servers now accept much larger backup uploads before saying a request is too large. When your AI provider has a hiccup, features now retry and recover instead of giving up, and the Skills helper tells you what to do instead of a blank error. And after Mem saves a link or note and replies, you can ask a follow-up right on the card. Nothing to do on your end, just update.

Added

Paseo is now documented as a supported orchestrator. Connect the AI tool Paseo launches for live capture, or run `nmem t sync --from paseo --all-projects --apply` to import older child sessions without relabeling them as Paseo.

Fixed

The "needs rebuild" prompt stops repeating after an upgrade. Moving up from an older build could leave Search asking to rebuild on every launch, even right after you rebuilt it; it now rebuilds once and settles.
A backend that can't start no longer drags down the whole computer. If startup keeps failing (say, a LAN-access setting carried over from an older version), the app now stops retrying after a few tries and shows a clear, retryable error, instead of quietly relaunching the backend over and over.
Asking Mem about a flag no longer quietly fails. A brief network hiccup as the answer was starting could leave the card stuck on "thinking"; it now retries the blip and always ends with an answer or a clear error you can retry.
Newer Apple Silicon Macs no longer crash the memory engine at startup. A change in Apple's latest Metal path could abort loading the embedding model on boot; Mem now avoids that path by default.
Contradiction and duplicate flags can keep the OLDER note as the current one. Next to "keep newer", a "keep older as latest" choice lets the earlier version stand when that's the one you trust.
Large imports finish. Bringing in a big history (thousands of messages at once) could run for a while and then lose the local server; large imports now run reliably all the way to the end.
Huge self-hosted backup uploads stop bouncing at 413. Docker installs now default to a much larger upload ceiling for ChatGPT, Alma, and ChatWise exports, and the server plus Caddy proxy use the same setting so you do not fix one layer only to fail at the next.
The Skills helper stops dead-ending. Creating a skill no longer fails with a blank "couldn't reach the model." It retries a passing hiccup, falls back to a model that's working, and if your own model provider rejects the request it says so and points you to switch models.
Access Anywhere reconnects on its own and is easier to diagnose. Before, if the tunnel dropped but its helper was still running, it could stay offline for hours while the app looked fine; Mem now watches the actual connection, brings it back automatically, defaults to a more firewall-friendly connection, and records the helper process in the settings file so support logs show what happened.
Gemini embeddings work again. A Gemini embedding model saved with a provider prefix was sent with a doubled name and rejected as "not found"; the id is now normalized so Gemini embeddings resolve.
Bringing in your own skills works. Importing a skill from a folder your AI tools already scan was rejected for everyone; it now reads the folder and brings the skill in (symlinked skill folders included).
The nmem CLI stops tripping up your agents. Search flags your AI was told to use (like -t week, plus --event-from/--event-to and --labels/--meta) now work instead of erroring out, and when a flag really is wrong the CLI returns a clear, machine-readable hint (JSON with --json) pointing at the right one, so an agent fixes it on the next step instead of stalling.
Linux users can get their Access Anywhere key again. The Rust nmem CLI has the top-level key command back, including rotate, login URL, and QR-code output for headless setup.
The nmem CLI reports failures instead of faking success. A command that failed (a delete that hit nothing, a provider test that could not connect, an ingest that could not read a file) now returns a real error exit code, so a script or agent notices and reacts instead of carrying on as if it worked.
Take ownership actually takes ownership. When you bring in one of your own hand-written skills and choose to take it over, Mem now moves the original aside while its managed copy is on (so your tool shows one skill, not two) and puts the original back when you turn it off. The original is only ever moved, never deleted.
Windows connectors stay out of the way. Claude Code, Codex, Copilot CLI, Kimi Code, Proma, Alma, Bub, and Hermes no longer flash a command window when their Mem plugin saves a session or reads context in the background.

Added

Record richer memories from the command line. nmem m add now takes the source conversation, message range, and time framing (--source-thread, --source-range, --when), and nmem m search can filter by when a memory was saved and include past versions (--recorded-from, --recorded-to, --include-history), matching the flags the guidance already teaches.
Type your own model id. The model picker now lets you enter a model that isn't in the fetched list (for embeddings and chat), so a custom or brand-new model id just works.
Paste an image even when your model can't see. If your chat model isn't a vision model, Mem now uses your configured vision model as its eyes: it reads the pasted screenshot, hands the text to the agent, and the agent works with it, so a pasted article gets recorded instead of coming back with "you didn't provide any content." With no vision model set up, it tells you clearly what to do instead.
Paste an image into the Timeline. Ctrl+V (or Cmd+V) a screenshot straight into the composer and ask about it; Mem sends it to the AI as something it can actually see, the same way the AI-Now chat already does.

Improved

AI features shrug off provider hiccups. When your model provider rate-limits or briefly fails, background AI (memory quality checks, link suggestions, and more) now retries and recovers instead of silently giving up.
Keep the conversation going from a saved card. After you drop a link or note and Mem replies, an Ask follow-up appears right on that card, so you can dig deeper without starting a new thread.
The Skills page types smoothly. Searching and browsing skills no longer stutters on longer lists.
Memories saved through MCP keep their source. If several tools write to the same Mem server, new memories can now show whether they came from Codex, Claude Code, Alma, or another client instead of all looking like plain mcp entries.

0.10.8

2026년 7월 2일

0.10.82026년 7월 2일

Fewer, Better Memories

This release is about signal over noise. Casual chats and passing agent chatter stop turning into memories, a new review drops throwaway notes before they are saved, and crystals stop piling up as near-duplicates. Your own standing rules now shape what gets captured. Claude Code session capture works on Windows again, self-hosted backups over Docker work end to end, and space-scoped graphs stay in their space. Nothing to do on your end, just update.

Improved

Fewer, better memories. Casual back-and-forth and an agent's passing working notes no longer get saved as memories, and a new durability review drops throwaway lines before they land, so your library keeps the decisions and lessons worth keeping and skips the noise.
Crystals stop piling up. When a synthesis already covers the same source memories, Mem reuses it instead of creating a near-duplicate, so the crystal list stays meaningful.
Your standing rules shape what gets captured. Active guidance rules are now applied while a conversation is distilled into memories, so capture follows the way you have told Mem to work.
Clearer memory counts. Stats and the Memories view now agree on the same Active, Archived, and All tiers everywhere you look.
The Knowledge Graph opens fuller. It starts with a larger, more representative sample, and small spaces fill the canvas instead of looking empty.

Fixed

Background capture no longer looks like it failed on long conversations. Saving a big session from Claude Code could time out and report nothing saved even though the work was fine; capture now gets the time it needs, so long threads save on the first try.
Self-hosted Docker backup and restore work end to end again. The built-in export and restore now run over the local API, which the new self-contained image needs, and the export carries your full graph content instead of empty rows.
Background distillation no longer hangs the app, and a conversation that gets distilled twice in quick succession is now handled once instead of producing duplicates.
Importing a lot at once no longer stalls the processing of your conversations into memories. A big batch used to hit an hourly limit meant for optional background work, then crawl for a long time while the queue piled up and looked stuck until a restart. That core processing now runs as fast as your AI allows, so a large import gets turned into memories promptly.
No more phantom spaces. Reading a project no longer quietly creates a space named after the folder that you never asked for.
Windows: Claude Code session capture works again. The Stop hook could report the session directory was not found and fail; the session-folder name is now computed to match Claude Code's own naming on Windows, so your sessions save automatically as before.
The Knowledge Graph stays in the space you're viewing. A space-scoped graph with only a few nodes no longer falls back to pulling in nodes from other spaces.

0.10.6

2026년 7월 1일

0.10.62026년 7월 1일

Skills, Tray, and Installers Back in Shape

A fast follow-up to 0.10.5. Skills work again: the app detects the AI tools you already have installed and shows each skill's full procedure. On Windows the app comes back from the tray instead of staying hidden, and it no longer gets stuck restarting after an interrupted rebuild. The nmem command line updates alongside the app again, and self-hosted RPM installs start cleanly. Nothing to do on your end, just update.

Fixed

Backup and export include your threads and their messages again. The export was writing an empty content store (0 documents, 0 messages) because it failed to capture the internal ids, so a restored backup lost all thread message content. Exports now contain the full content store, and the same bug that made graph node exports come out empty is fixed too.
macOS: quitting from the Dock no longer hangs. Dock → Quit used to freeze with a spinning cursor and skip the usual quit screen while it shut down; it now uses the same clean quit flow as Cmd+Q.
The desktop app no longer gets stuck re-prompting you to rebuild search. On a slower machine a long on-device rebuild could be cut short by the app's own health check before it finished, so the rebuild never completed and it kept asking again on every restart. The app now waits out an in-progress rebuild instead of restarting the engine underneath it.
Self-hosted / headless: your web-login key is easy to find again. On a headless server the backend now makes sure a login key exists and prints it to the log at startup (matching the earlier behavior), so you can copy it to sign in to the web UI.
Skills detects your installed AI tools again. On 0.10.5 every tool showed as "not detected" (even ones you had set up) because the app looked under the wrong home folder. It finds them now, so you can connect your skills to Hermes, Claude Code, Codex, OpenClaw and the rest.
Every skill shows its full procedure again. Compiled skills were rendering with an empty body; the app now reads the skill's content so you can see and copy exactly what it does.
Windows: the app comes back from the tray. After minimizing to the tray the window could stay hidden and only a restart brought it back. Clicking the tray icon now reliably restores and focuses the window.
Windows: the app no longer gets stuck restarting after an interrupted rebuild. If restoring your Access Anywhere tunnel crashed the backend on startup, the app would loop forever. It now detects that and starts without the tunnel, so you can get back in and re-enable it from Settings.
The nmem command line updates alongside the app again. Some setups hit "nmem_cli package not found" and the CLI stayed on the old version. The bundled nmem now tracks the app, and a stale wrapper left by an older version is repaired on launch.
Self-hosted: RPM installs start again. Since 0.10.3 the .rpm was missing the backend binary and failed to launch. The RPM now bundles it, so it starts like the .deb and Docker builds.

0.10.5

2026년 6월 30일

0.10.52026년 6월 30일

Steadier on the New Engine

A focused round on the new engine, with a couple of welcome additions. Reminders you ask the assistant for now actually fire and land on your Timeline, background processing stays calm even on a slower AI, the activity calendar shows the right day for your timezone, and switching spaces now filters your Threads too. Rebuilding search also reads clearly now: a finished rebuild restarts the engine to free memory (normal), and rebuilding right after a model change waits out the restart instead of flashing an error. On top of that, the things a few people ran into on 0.10 are sorted: rebuilding on-device search runs to the end, your Pro license and settings stay put, and connecting your own cloud AI provider works start to finish. Nothing to do on your end, just update.

Added

Ask the assistant to remind you about something on a future day, and it now actually does. "Remind me next Monday to follow up on the database decision" schedules a real reminder that surfaces on your Timeline when the day comes, instead of being acknowledged and quietly forgotten.
Pi now starts every session already knowing your context. Connect Pi to Nowledge Mem and it pulls in your working memory and recent context the moment it launches, so it picks up where you left off instead of starting from a blank slate.

Improved

Background processing stays calm even when your AI is slow. If you point Mem at a slower provider, background work used to pile into hundreds of queued tasks and burn tokens without ever catching up. Mem now keeps that queue bounded and lets its regular catch-up passes handle the rest, so it stops stacking and stops wasting tokens.
Rebuilding search reads clearly now. When a rebuild finishes, Mem restarts the engine to free up memory — that's normal, and it now says so instead of looking like a crash. And if you rebuild right after switching your embedding model (which briefly restarts the engine), it waits out the restart and retries instead of flashing a connection error.

Fixed

Connecting an AI tool to Mem over your network works again. Once you opened Mem to other devices on your Wi-Fi, or to Access Anywhere, pointing a tool's MCP connection at it could fail with "Host header is not allowed". Connections now work whether the tool reaches Mem on this computer or across your network. (Tools pointed at 127.0.0.1 on this computer were never affected, and that stays the right address for a local setup.)
Switching spaces now filters your Threads too. The Threads list ignored the space switcher and showed threads from every space at once; now it shows just the selected space's threads, the way Memories already did.
The activity calendar and Timeline now show the right day for your timezone. They used to group everything by UTC, so a few timezones away your activity looked shifted by a day and clicking a date could come up empty. Now everything is grouped by your local day — events appear on the day they happened for you, and picking a date shows that day's activity.
A Daily Briefing you re-run by hand now shows on the Timeline. If the briefing failed to run on its own and you triggered it again yourself, the result didn't always appear. One you ask for now reliably lands there.
The assistant's rule suggestions run again. Mem quietly reviews your memories for durable working rules worth keeping and proposes them as AI Context rules, but that background pass had started failing every time, and the manual "Run" did nothing. Both work now, so suggested rules show up again.
Kimi's Coding Plan works on both the China and global endpoints. Whether you point it at the mainland-China or the global address, chat now connects correctly instead of erroring out.
Rebuilding search on a local model finishes. If you moved to the new on-device embedding model and the rebuild spun the fan forever with the progress bar stuck, it now runs all the way through and your semantic search comes back.
Self-hosted: search rebuilds work with any embedding model, not just the default size. If you point Mem at your own embedding provider (OpenAI, OpenRouter and the like) whose vectors aren't the default width, rebuilding used to fail with a dimension-mismatch error. Mem now sizes the search tables to your model automatically, so rebuild and semantic search work end to end.
A search rebuild that can't make progress now reports a clear failure instead of spinning forever. If the embedding step keeps erroring, the rebuild stops with an error you can act on, rather than sitting at 0% indefinitely.
A successful search rebuild now stays done — it stops asking to rebuild on every restart. After a rebuild finished, some setups kept showing "Rebuild needed" again after each restart, and reinstalling didn't help, because the app forgot to record which model it had just rebuilt with and re-decided it was out of date. It now remembers, so once a rebuild completes the prompt clears and stays cleared.
Your Pro license, your settings, and the AI providers you've saved all stay put — even on Windows, where the app used to read them from a different folder than it saved them to. That mismatch could make a lifetime license look lost, a background-AI budget refuse to change, or a custom AI provider (and its key) disappear after an update. The app now keeps everything in one place, and brings forward anything an earlier version left behind.
Connecting your own AI provider works end to end — for chat and for embeddings. The address you enter is accepted even without a /v1 at the end, so it works at runtime and not just on the test; embedding-model lists load again for gateways like OpenRouter; and a provider that doesn't pass the test can still be saved. OpenAI-compatible services such as Synthetic and SiliconFlow now connect cleanly, and you're never stuck between chat and embedding.
The built-in AI's tools work even with a network proxy running. The assistant was trying to reach itself at an address some proxies reject outright; it now uses a local address that always works, so its memory and graph tools come back.
Self-hosted: the docker control script works with the new engine image again. Status, start, restart, license and key commands now talk to the running server directly, instead of relying on tools the slimmer image no longer ships, so they stop reporting false timeouts.
Self-hosted Linux: the `nmem serve` service starts again. If you run Mem as a systemd service, upgrading no longer leaves the unit in a restart loop — `nmem serve` now launches the bundled engine, so your service comes back up after the update. Your data was never affected.
Windows: your memories are no longer "missing" after upgrading from an older version. If your data lived in the older Windows location (the roaming AppData folder), the new engine looked in the current location and showed an empty library. Mem now finds and opens your existing graph from the old location automatically. Your data was always safe; it just wasn't being read.
Self-hosted: heavy moments no longer make the container restart itself. On a small server, asking the assistant a question while background processing was busy could make the health check time out, so the container marked itself unhealthy and restarted mid-work — looking like a crash. Mem now keeps enough threads free to always answer the health check, and the check itself is more patient, so a busy server stays up.
Updating restarts the app, and the command-line tool keeps up. On Mac an update could finish installing without reopening the app, which also left the nmem command-line tool on the old version. The updater now relaunches the exact app you were running, and the CLI refreshes to match the next time the app starts.

0.10.3

2026년 6월 29일

0.10.32026년 6월 29일

Settling In the New Engine

0.10 moved Mem onto a new native engine. This update smooths the rough edges a few people hit right after that change: hosted embeddings, rebuilding search, the update prompt, and the self-hosted web app on every image. Nothing to do on your end, just update.

Fixed

Hosted embedding models work again. If your semantic search runs on a provider like Jina (including Nowledge AI), saving and searching no longer fail; the request is now sent in the exact shape those providers expect.
Rebuilding search is solid from start to finish. After switching or updating an on-device model, a rebuild used to stall or look like it failed, and the "Rebuild Needed" badge would linger even when everything had actually reindexed. Now the new model loads, the rebuild runs to completion, and the badge clears the moment it succeeds.
No more phantom update prompt. After you update, the app no longer keeps offering you the version you already have. The update check now reads the installed app's real version.
Self-hosted Docker serves the web app on every image. Opening Mem in the browser at /app works on the standard image and on the GPU images (NVIDIA CUDA and AMD/Intel Vulkan) alike; the web client is now bundled in all of them instead of returning a 404.

0.10.0

2026년 6월 29일

0.10.02026년 6월 29일

A Rebuilt Engine: Faster, Lighter, Everywhere

Nowledge Mem now runs on a new native engine. The app starts faster, uses far less memory, and the macOS download drops from over a gigabyte to about 150 MB (around 350 MB once installed). Everything you saved stays exactly where it is, with nothing to migrate.

Highlights

A new native engine: faster startup, much lower memory, and a far smaller download. The macOS installer goes from over 1 GB to about 150 MB, and installed size from roughly 1.3 GB to about 350 MB. Nothing to migrate; all your memories stay in place.
See which AI tools are using your memory. Integrations now shows live MCP traffic, so you can watch the calls your connected tools make in real time.
Read text out of images with your own model. Library OCR now works with your local or custom vision models, so scanned PDFs and screenshots get understood on your terms.
On-device work can use your GPU. Apple Silicon (Metal) is automatic, with nothing to set up. For GPU on Windows or Linux, self-host with Docker: NVIDIA (CUDA) and AMD or Intel (Vulkan) images are available.

Improved

Local search and on-device models now run natively on Windows and Linux, not just Mac, so meaning-based search works the same on every platform.
Meaning-based search is more reliable: results match what you actually meant more often, even when your wording differs from what you saved.
Background processing of synced conversations is smoother. New conversations are worked through in shorter, steady waves instead of stalling behind long quiet windows after a large sync.
PDFs come in with their structure intact. Tables, headings, and lists are rebuilt as clean Markdown instead of one flat block of text, so what you save from a document reads the way it looked.
Self-hosting with Docker now includes Access Anywhere out of the box: the Cloudflare tunnel is built into the image, so you can reach your own server remotely without running a separate sidecar.

Fixed

Linux desktop startup is safer on native Wayland desktops. Mem no longer forces XWayland by default, while users who prefer it can still opt in with NOWLEDGE_FORCE_X11=1.

0.9.29

2026년 6월 29일

0.9.292026년 6월 29일

Smoother Background Processing

0.9.29 keeps background conversation processing moving after large sync backlogs and makes Library image OCR work with user-configured vision providers.

Highlights

Background processing now drains newly synced conversations in shorter, bounded waves instead of leaving them behind long quiet windows. Existing one-hour pending states are migrated automatically on startup.
Library OCR now trusts custom and local vision models when you choose “Same as default,” so OpenAI-compatible, Ollama, and LM Studio vision providers can recover image text without being falsely blocked.

Fixed

Automatic thread analysis now recovers from large backlogs and full worker queues without making users wait behind stale one-hour timers.
Knowledge Graph backfill now skips memories that are no longer visible instead of getting stuck retrying them, so graph maintenance can keep moving after cleanup or lifecycle changes.
Returning from a focused graph search to the full Knowledge Graph now resets the old search context, so the search field, layer controls, and assistant panel do not describe a focus map while the full graph is on screen.
Library image-text recovery now works with user-controlled custom vision providers unless they explicitly opt out of vision support.
Skill suggestions now check your existing local Skills before suggesting a new one. If you already have a hand-written Skill for that workflow, Mem avoids creating a weaker duplicate and points you toward importing the original instead.
Linux desktop startup is safer on native Wayland compositors. Mem no longer forces XWayland by default, while users who prefer XWayland can still opt in with NOWLEDGE_FORCE_X11=1.
Closing the desktop window is more reliable. The window no longer disappears and then pops back once before staying closed.

0.9.28

2026년 6월 29일

0.9.282026년 6월 29일

A Clearer Knowledge Graph Exploration

0.9.28 makes Knowledge Graph search easier to read and turns graph exploration into a clearer path: start from the subject, understand the evidence, then keep thinking with the graph assistant.

Highlights

Graph search now opens a quieter exploration view, separating key matches, supporting evidence, and surrounding context instead of showing every nearby object with the same weight.
Selecting a node or connection now shows a structured Details view with what it is, why it appears in the current map, nearby objects, evidence, and useful actions. You can click nearby objects or ask Mem to interpret the selected item, rather than landing in a raw inspector.
Library Wiki summaries are stricter: fallback text no longer pretends to be an AI-written Wiki summary, and new topic summaries are written more like compact Wiki leads.

Improved

Knowledge Graph exploration now keeps the map usable while the assistant interprets it, with a calmer View control for Key, Context, and Whole levels of visual detail.
Compound graph searches now feel less like exact-node lookup. You can search a phrase or question such as “AI and biology”; Mem expands it into grounded graph handles while keeping your original query as the subject of the exploration.
Graph interpretation is easier to continue: follow-up actions and Details actions are tied to the current focus instead of generic buttons, and returning to Graph no longer replays the same automatic reading.

Fixed

AI Now adapts automatically when a custom model gateway rejects reasoning parameters.
Desktop quit is more reliable when the shutdown screen cannot finish, so users should not need to force-quit Mem just to close it.
Windows minimize and maximize buttons are more reliable on the custom title bar.
The native “open in separate window” workspace button is hidden on Windows for now, preventing a Windows-only title bar freeze while the multi-window path is being hardened.
Library Wiki no longer counts structural fallback text as an AI summary, so old low-quality topic cards can be refreshed instead of staying mislabeled.

0.9.27

2026년 6월 27일

0.9.272026년 6월 27일

Library Tags, Clearer Setup, and Better Ollama Support

This patch fixes Library tag selection, makes first-run AI setup clearer, and improves Background Intelligence reliability with local Ollama models.

Highlights

Library tags now appear selected correctly and update immediately when you add or remove them.
Fresh installs now explain that the built-in local model still works for local search and basic parsing, while Timeline chat, AI Now, and Background Intelligence need Nowledge AI or your own provider.
Local Ollama models are less likely to fail with empty responses during structured Background Intelligence tasks.

Fixed

Library tags use the same IDs across source lists, search, and detail panels, fixing missing existing tags and failed toggles in the desktop client.
Background Intelligence setup errors now point users to Settings → Providers instead of showing raw 503 messages.
Ollama models are more reliable for structured background tasks, especially local models that do not handle Ollama JSON mode consistently.

0.9.26

2026년 6월 26일

0.9.262026년 6월 26일

Quieter Timeline and Background Memory Fixes

This release makes background work quieter, keeps synced conversations moving, and fixes a case where distillation could run without saving Memories.

NotebookLM conversation capture in Nowledge Mem Exchange — Exchange now recognizes NotebookLM as its own source when you save the visible notebook chat.

Highlights

Timeline now groups routine background noise into a quiet daily digest while keeping failures, review requests, and created knowledge visible.
Fresh synced conversations no longer wait behind longer automatic distillation work before their title and summary are reviewed.
Exchange can now capture NotebookLM chats and keeps Doubao conversations complete after Doubao's latest UI update.

Improved

Routine background updates are grouped instead of filling Timeline with dozens of empty-looking events. Actionable updates still stay visible.
Conversation sync is fairer under load: lightweight title and summary reviews can finish before longer automatic Memory distillation work.

Platform

Fixed

Synced conversations can create long-term Memories again even if the extra validation check is temporarily unavailable. Mem still blocks obvious template leaks and duplicates, but a validator outage no longer makes background distillation look successful while saving nothing.
Background distillation history now shows when a proposed Memory was skipped by a quality gate, so “the task ran but no Memory appeared” has a visible reason.
Context rules and Skills are easier to review: active items stay first, suggestions are folded, and manually added rules no longer risk replacing another rule with the same ID.
OpenClaw connector 0.8.27 works with newer OpenClaw builds that require plugin tools to be declared before startup, so OpenClaw conversations can keep syncing after the OpenClaw update.

0.9.25

2026년 6월 25일

0.9.252026년 6월 25일

Small Reliability Fixes

A small patch for AI Now, Windows, and the browser extension.

Highlights

AI Now, Windows desktop, and the browser extension each get a small reliability polish.

Fixed

AI Now Quote now keeps the selected text visible in your message.
Windows no longer flashes helper cmd windows during background work.
Exchange keeps edited ChatGPT turns safer, repairs unfinished Gemini replies, and shows clearer standby/offline states.

0.9.24

2026년 6월 25일

0.9.242026년 6월 25일

Library Wiki Topic Pages

This release improves how Library Wiki topics are written, so generated topic pages read like grounded notes about your work instead of graph-processing reports.

Highlights

Fixed

Wiki topic summaries now describe the subject directly instead of saying things like “this community”, “this document”, or “this article”. Existing degraded summaries are detected and refreshed first.
Library extraction now writes and repairs provenance links between extracted memories and their original documents, so documents no longer appear as disconnected graph islands after their memories are created.

0.9.23

2026년 6월 24일

0.9.232026년 6월 24일

Memory Health Catch-Up and Kimi Work

This release helps larger memory libraries stay healthy as they grow, adds Kimi Work support, and tightens the identity and provider paths used by advanced agents.

Memory Health review showing maintenance pace and review coverage in Nowledge Mem — Memory Health now explains the review pace, recent coverage, and backlog before you decide what to tidy.

Library Wiki export showing Open Knowledge Format as a portable knowledge bundle — Open Knowledge Format exports turn the Library Wiki into a portable, vendor-neutral knowledge bundle.

Highlights

Platform

Improved

Fixed

0.9.22

2026년 6월 23일

0.9.222026년 6월 23일

Maintenance Previews and Setup Fixes

Fixed

ChatGPT Web / Remote MCP setup can now be restarted immediately if the pairing or approval tab was closed, without waiting for the 30-minute pairing window to expire.
Maintenance previews remain inspectable even before Background AI is configured. You can still see what Memory Maintenance, Wiki Topic Detection, or Knowledge Graph backfill would review; only running the work needs the background agent.

0.9.21

2026년 6월 22일

0.9.212026년 6월 22일

Cleaner Wiki Topics and More Resilient Background Work

This release makes Library Wiki topics more selective, keeps slow topic-summary calls from failing the whole refresh, and smooths several daily app flows.

Highlights

Library Wiki topics now require real repeated memory evidence, so orphan entities and one-off file/path names no longer become noisy topic pages.
Wiki topic refreshes now survive slow model calls: Mem keeps the refreshed topic graph and fills remaining AI summaries in later bounded waves.
Skill update flows, desktop window restore, and the bundled nmem TUI all received reliability fixes.

Fixed

Library Wiki topics now require enough memory-backed evidence, keeping weak graph buckets, orphan entities, and one-off file/path names out of user-facing topic pages.
Slow AI summary calls are now bounded during Wiki topic refresh. Mem keeps the refreshed topic graph, records partial summary work clearly, and lets later enrichment waves fill the remaining pages.
Skill update buttons no longer freeze and jump back, desktop windows restore across restarts, and bundled server installs can launch nmem tui.

0.9.20

2026년 6월 22일

0.9.202026년 6월 22일

AI Now Skills, Safer Exports, and Clearer Reviews

This release tightens the places where Mem connects daily work to long-term knowledge: creating skills from AI Now, backing up large servers, reviewing background work, and resolving Feed conflicts.

Highlights

Added

AI Now has a first-class Skill creation path. Use the slash menu or ask AI Now to create a skill, and Mem will draft a reviewable Skill with the normal Skills lifecycle instead of creating a memory note about the request.

Improved

Skill Suggestions keep moving when one review pass reaches Mem's safe evidence-reading guardrail. Mem queues a small follow-up pass that skips evidence already reviewed, so a large model context window is no longer confused with an unbounded review.
Processing history now shows Mem's own per-run safety caps as protected partial reviews, not red failures. It is clearer when a task paused intentionally, when a provider failed, and when a user-configured token budget stopped the run.
AI Now file previews are smoother for generated work. HTML folders can load sibling CSS, JavaScript, images, and other assets; Markdown and HTML can pop out into a separate preview window; desktop users can also open or reveal task files locally.
Feed conflict reviews now show real choices in the detail panel. Keep newer, keep both, or no action are presented as actions instead of a confusing sentence of suggested labels.

Fixed

Platform

The Community integration catalog was refreshed so packaged integrations and marketplace metadata match this desktop release.

0.9.19

2026년 6월 21일

0.9.192026년 6월 21일

Recoverable Background Tasks and Safer Skill Workflows

Fixed

Platform

0.9.18

2026년 6월 20일

0.9.182026년 6월 20일

Safer Memory Maintenance and Cleaner Wiki Topics

Improved

Fixed

AI Now is more compatible with Responses API providers that expect instruction prompts as developer messages instead of system messages.
AI Now refreshes recoverable MCP OAuth credentials before asking you to reconnect, so connectors such as Notion should prompt less often.
Library Wiki topic refreshes now self-repair their background job tracking after an upgrade or restore. If the local database was missing that tracker, Wiki clustering starts normally instead of showing an internal service error.

0.9.17

2026년 6월 19일

0.9.172026년 6월 19일

Better Memory Health and Search Gates

Improved

Fixed

0.9.16

2026년 6월 18일

0.9.162026년 6월 18일

Safer Upgrades and Clearer Background Sync

Added

Grok Build now has a dedicated Nowledge Mem connector. New Grok sessions can start with Mem context, save real Grok transcripts through hooks, and backfill older local sessions with `nmem t sync --from grok`.

Improved

Newly synced small conversations now enter background analysis much sooner, while large historical threads still wait and drain in bounded waves. This restores the normal “save a useful conversation and Mem quietly learns from it” path without turning bulk imports into surprise model spend.
Task Center now shows when newly synced conversations are waiting for the quiet window before analysis, so background sync no longer looks idle while Mem is intentionally waiting for the conversation to settle.
Memory Health reviews now explain the real choices more clearly: mark memories as still useful, hide stale low-risk facts and events from everyday recall, or run a bounded related-memory review. Mem still does not delete, rewrite, hide, or merge anything unless you choose an action.
Memory Health now uses one review plan for its pressure, next step, and available choices, so Stats, Timeline, and Processing explain the same maintenance state instead of giving conflicting signals.
Memory Health reviews now reuse an unresolved Timeline review when the same batch is still waiting for your decision, instead of creating duplicate review items on each scheduled pass.
Timeline now shows when Mem finishes distilling long-term memories from a synced conversation, including whether new memories were saved and where they came from.

Fixed

Failed link captures no longer look like saved Library items. When Mem cannot read a page through Browse Now or the basic HTTP fallback, Timeline now says the link was not captured, explains that it is not searchable yet, and lets you retry after Chrome and the Nowledge Mem extension are connected.
Single conversations synced by integrations are no longer mistaken for large historical imports. They now enter the normal settling-window analysis and bounded memory distillation path unless the integration explicitly marks them as historical or cost-controlled.
Queued conversation distillation now survives app restarts and hot reloads. Interrupted follow-ups retry instead of disappearing before Mem has actually learned from the conversation.
Upgrade-time search maintenance no longer gets stuck on “Updating metadata”. If the lightweight repair cannot finish safely, Mem now switches to the normal Search Rebuild path instead of spinning in the background.

0.9.15

2026년 6월 17일

0.9.152026년 6월 17일

Better Wiki Summaries and Maintenance Reviews

Improved

Memory Maintenance cleanup reviews now appear in Feed Timeline when they need a decision. The review card shows the current batch, explains what Mem will do, and keeps the safest default: finding grounded relationships without deleting, rewriting, hiding, or silently merging memories.
Memory Maintenance now runs a weekly read-only health review by default. It checks a bounded slice of memory freshness data, does not call a model, and only asks for attention when a Timeline decision card is useful.
Stats now shows Memory Maintenance pressure instead of only showing how fast your memory library is growing. You can see healthy, needs-review, and already-hidden counts, then prepare the Timeline review card from the same place when action is needed.
Memory Compaction previews stay fast by default. The expensive recent-duplicate semantic probe is now opt-in, so opening a preview does not unexpectedly wait on remote embeddings.
Library Wiki topics can now fill in missing AI summaries without rebuilding the whole topic graph. If your topics are already fresh, Mem can write the missing summaries as a bounded follow-up pass instead of rerunning community detection.
Library now shows when Wiki pages still need AI-written summaries and lets you start that bounded summary-writing pass from the Wiki itself.
Topic refresh review now separates graph freshness from summary coverage, so Processing can show when new memories need a topic refresh versus when existing topics only need missing Wiki summaries.
Before writing AI summaries for Wiki topics, Processing now shows an estimated token cost, so large libraries are easier to review before spending managed AI credits or BYOK usage.
Working Memory refresh is stricter about startup context. Event-driven updates must cite the memory that triggered them, and they cannot silently remove existing Needs Attention context without marking the flag as resolved.
Daily Briefing is stricter about evidence too. A briefing now needs Memory sources from the prepared context, and any Working Memory update it writes must link back to those sources.
Memory Maintenance cleanup reviews now focus on items that need a decision. Memories already hidden from everyday recall are still counted in diagnostics, but they no longer fill the review list.
Library search reliability checks now include your real imported sources: before release, Mem verifies that sampled Library items can be found by their own visible titles.
Learned Rules now have a stricter release check. Mem verifies that active learned rules still have readable supporting evidence before they can shape an AI's startup context.
Memory Maintenance cleanup reviews now show a balanced ledger: what is still active in everyday recall, what needs your decision, what you explicitly kept, and what is already hidden. That makes cleanup progress visible without making Mem more aggressive by default.
Bulk thread distillation plans are easier to trust. Mem now separates the first safe wave from later, too-large, and source-repair groups, so large imports stay searchable without turning into a surprise AI workload.
Background maintenance previews are more trustworthy before they reach you. Mem now checks label cleanup and skill maintenance against the running library in read-only mode, with safeguards for missing model dependencies, accidental writes, and oversized previews.
Release checks now exercise real preview paths for Knowledge Graph extraction and thread distillation in read-only mode. This helps catch provider or preview-shape issues before they reach a user, without creating memories or graph changes during the check.
Background Knowledge Graph extraction is now more resilient. Mem reads only the selected memory wave, decides what graph facts are grounded, and writes them through a deterministic graph-write tool instead of relying on one brittle extraction pass.
Deferred thread distillation now works more like a careful review. Large imported or selected conversations are searched and sampled first, then only high-signal memories are created with thread provenance attached.
Memory Link search now understands relation intent better, including relation labels and phrases such as memory links, same-topic links, and related memories, while still requiring endpoint evidence before relation candidates affect retrieval.
Overview source metrics now make it clearer what has been imported, parsed, and made searchable in Library.

Fixed

Conversation-summary review now restores the daily automatic Memory path. Live or incremental synced threads still wait for a quiet window and update title/summary first, but when that review finds durable knowledge it schedules a bounded agentic distillation follow-up. Large historical imports remain searchable immediately and still require an explicit distillation plan before spending model usage.
Batch conversation imports now leave one Timeline handoff instead of firing analysis for every imported thread. The card can queue the next small memory-distillation wave, while the imported conversations stay searchable immediately.
Scheduled Memory Maintenance cleanup reviews now respect the cleanup setting. Turning stale-memory review off keeps the scheduled review quiet, while manual cleanup reviews remain available.
KFS time-lens label folders now paginate after server-side label aggregation, so large libraries do not hide active labels near the end of the graph.
Label cleanup live dry-run exports are less brittle in release checks when a preview reports total planned merges but omits a separate canonical-merge count.
Background intelligence tasks no longer fail just because they reached a per-task reading budget. When Mem has enough evidence, memory and thread tools now return a clear stop signal so the task can finish cleanly instead of retrying the same blocked read.
Background tasks started during backend warm-up no longer turn into a vague service-unavailable failure or a false success. Mem records them as not-started, keeps the diagnostic detail, and retries after a short pause.
Source import now handles Org-mode files and standalone images more reliably, including image-only sources that need vision OCR.
AI Now is smoother on long timelines: message virtualization and the thread minimap are less likely to jump while you navigate.
AI Now generated files are easier to use from web and remote clients. Text, Markdown, and code artifacts can open in a right-side preview, and task files can be browsed or downloaded from the session.
AI Now reports can now be saved into Library as durable artifacts from the preview panel. Generated Markdown, text, and HTML outputs become searchable Library knowledge instead of staying trapped in a temporary task folder.
Exchange now explains Mem API authorization failures more clearly instead of showing a generic connection problem.
The AI-summary limit selected in the desktop app is now passed all the way to the backend when refreshing topics.
Desktop Working Memory refresh now respects the selected space, matching the web and remote clients.
Thread distillation previews can now use explicit imported conversation text without first requiring that conversation to exist as a stored thread. Stored conversations still keep their normal provenance links.
The opt-in NAS first-run key reveal is stricter about forwarded IP headers. Mem now trusts those headers only from a local or private proxy, so a public caller cannot spoof a LAN address.
Memory Maintenance cleanup review now uses freshly computed freshness before trusting cached decay scores, so recently used memories are not shown for cleanup because of stale cache data.
Search metadata updates after an upgrade now finish during the same startup warm-up that detects them. The title bar should no longer stay on “Updating metadata” until a second restart, and older LanceDB builds can complete the metadata update without a full rebuild or re-embedding.

0.9.14

2026년 6월 16일

0.9.142026년 6월 16일

Clearer Memory Maintenance

Added

Memory Maintenance now has a cleanup review. Mem separates memories that are safe to review for reversible archive, memories that should go through semantic compaction first, and memories already hidden from everyday recall. The review does not archive, merge, or delete anything by itself.
Cleanup review now leaves explicitly kept memories alone. Pinned memories and memories marked to keep are not shown as cleanup candidates just because they are old.
Cleanup review now explains the whole scan, not just the candidates: which memories look healthy, which are explicitly kept, and which are already out of everyday recall.
You can now archive the safe candidates shown in cleanup review. Mem checks each memory again before applying the action, so anything pinned, recently changed, or no longer safe is skipped.
Cleanup review can also send the semantic-review items into Memory Compaction. Mem rechecks them first, then queues a bounded review; nothing is merged unless grounded duplicates are found.
Memory Compaction results are clearer in Feed Timeline. Compaction cards now show reviewed candidates, touched memories, relation changes, created crystals, review flags, and related graph context instead of only a generic agent summary.

Fixed

Reversible archive actions now update the graph lifecycle state correctly, so archived memories are hidden from everyday recall without being deleted.
Nowledge Brain chat history is safer when Graph opens. The panel now waits until saved history has been read before it writes any automatic save, preventing an empty first render from replacing the last conversation.
Wiki topic summaries are less likely to fall back to plain graph structure. When a topic has representative memories or entities, Mem can use that evidence for an AI-written summary even if the graph has not written many relationship edges for that topic yet.
Knowledge Graph Extraction task history now shows estimated model usage when a provider performs extraction but omits token accounting. Successful extraction runs should no longer appear as “No model call.”
Text model calls from OpenAI-compatible, LiteLLM, ChatGPT Subscription, and Copilot-backed providers now fall back to estimated usage when the provider does not report token counts. Embedding usage is not mixed into this LLM task history.
Processing history now explains why a background run used no model tokens. It can show that there was nothing to process, no review was needed, a task was deterministic, the model was not configured, or the provider did not report usage.
Google Activity HTML imports are more tolerant of direct exports and renamed Takeout files, so common downloaded filenames are recognized without manual cleanup.
Thread Detail is more consistent with your memories. If a conversation already produced memories but an older run missed the thread link, Mem can still show those memories and their entities from exact provenance instead of saying nothing has been distilled.
Memory Knowledge Graph extraction now reports remote-model failures more honestly. Empty responses from a remote provider no longer look like a successful extraction with zero entities.
Desktop release packages are checked more deeply before shipping. Mem now verifies key model, search, local embedding, document parsing, and language-processing paths in the bundled runtime, catching missing dependencies before they reach users.

0.9.13

2026년 6월 16일

0.9.132026년 6월 16일

Safer Background Work and Credit Accounting

Improved

Mem's release checks now look at live Knowledge Graph coverage, thread distillation coverage, topic freshness, and topic refresh plans. That helps catch expensive or stale background-work patterns before they reach your library.
Background task audits now verify that every scheduled task points to real evaluation coverage. If a quality check is renamed or removed, release checks fail instead of silently trusting stale documentation.
Evolution follow-up work is now bounded. When several memories become linked by EVOLVES relationships, Mem checks for useful crystals in small waves instead of opening one large graph exploration.
Large conversation analysis is more restrained. Mem now gives background thread analysis a hard reading budget and shorter evidence excerpts, so a very long imported or coding session is sampled deliberately instead of letting one task grow without bound.
Scheduled follow-ups are now treated as lightweight progress checks. They can look at focused evidence, update Working Memory, or create one concise insight, but they no longer run broad graph jobs or open-ended background research.
Working Memory refresh is now narrower and more predictable. Saving a new memory can still update today's active context, but the refresh no longer branches into threads, sources, communities, or graph maintenance.
Recurring background reviews are more predictable. Morning briefings, insight checks, crystal reviews, and memory compaction now each run in their own bounded lane, so a routine review cannot drift into unrelated graph or source work.
Proma is now a first-class integration surface. Mem can show Proma conversations with the right source branding, treat Proma as a Skills host, and work with the Proma connector for startup context, session capture, and standard memory skills.

Fixed

Thread distillation and thread analysis no longer show a hard failure just because the background agent used all of its allowed steps. Mem keeps completed work, cools down the retry, and explains the old error more clearly in task history.
Manual thread distillation is now a bounded background lane. Large conversations are read progressively with thread-focused tools instead of inheriting unrelated graph, source, or review actions.
Older pending follow-ups now pick up a 16-step cap and a narrow evidence-reading profile when they run, so a routine reminder cannot unexpectedly consume a large amount of model credit.
Event-driven Working Memory refreshes now pick up a 12-step cap and a narrow read/write profile before they enter the queue, so routine memory saves are less likely to create expensive background runs.
Task history now keeps the contract for bounded recurring reviews, making it easier to understand why a background run stopped and which review lane it belonged to.
Managed AI credit settlement is safer after a gateway restart. Older pending usage records keep their settlement state when the gateway reloads them, reducing the risk of duplicate or lost credit accounting.
Plus account sign-in is more reliable from desktop, web, and remote clients. Sign-in link requests now match the account service format, backend account requests identify themselves clearly, and rate limits show a wait-and-retry message instead of a generic failure.
Managed AI and OpenAI-compatible providers are more reliable on a fresh desktop install. Manual distillation, background extraction, and remote embedding indexing no longer fail because the release package is missing the runtime client those providers need.

0.9.12

2026년 6월 15일

0.9.122026년 6월 15일

Cleaner Topics, Safer Rules, and Sharper Search

Improved

Search quality checks now protect multilingual queries without relying on an English stop-word list. Short words, CJK terms, accented text, Arabic text, and structured identifiers stay eligible, while more specific anchors still rank higher.
Memory Links now help search. When your query is really about how two memories relate, active links you created can help Mem bring the connected memories into the result set.
Library Wiki no longer depends on expensive AI-written topic summaries to show topics. A graph-only refresh can still show grounded topic names from the entities and memories inside each cluster.
The Library Wiki tab now explains where topic pages come from and can refresh topics directly. High-quality Wiki refreshes write AI summaries by default; structure-only refresh remains available when you explicitly want the low-cost path.
Background status checks are lighter. Common views can tell whether Mem is processing something without loading full task history, usage rollups, or graph diagnostics every time.
When AI topic summaries are enabled, Mem now writes them for the largest topic groups first instead of trying to summarize every topic in a large library at once.
AI topic-summary credits now go only to topics you can actually see. Mem's internal unassigned graph bucket no longer consumes the first summary slot or enters Library search as a real Wiki topic.
Mem's release checks now look at live Library topic names and summaries, not only whether background tasks finished. That makes weak topic names, stale community projections, and ungrounded summaries easier to catch before release.

Fixed

Desktop updates keep working after you enable Access Anywhere or connect the app to another Mem server. The installed macOS, Windows, and Linux apps update themselves through the desktop updater; if the connected server is a Docker/headless deployment, Settings can show it as a separate server update card, with an explicit server-local switch for allowing remote server upgrades.
Library Wiki and Knowledge Files no longer show Mem's internal unassigned community bucket as a real topic. The topic tree now keeps real topics, including topic id 0, and filters only the internal -1 bucket.
Rejected or dismissed Rules now stay out of agent startup context the same way draft and archived Rules do, across Context Bundle, CLI/KFS context, and agent-scoped profiles.
Fresh macOS and Linux desktop installs now finish `nmem` CLI setup in the background after the app is ready, so new terminal sessions can find the command without a separate repair step in Settings.
Connected MCP clients are more forgiving when a tool keeps sending an old space header after you renamed or removed a space. Explicit space choices still stay strict; only stale ambient context falls back safely.
Rules review is steadier when a model returns an empty judgment. Mem retries once, then reports a real provider failure instead of silently dropping the run.
Strict OpenAI Responses-compatible local servers now receive the message content shape they expect, which fixes provider validation and background distillation on servers such as Lemonade.
The Codex connector's Stop hook is more reliable on Windows and mixed shell environments. It now launches through a cross-platform helper and uses the plugin root passed by the host instead of depending on fragile quoting or PATH behavior.
Stopping Wiki Topic Detection now also stops the underlying graph job, and restarted apps clean up stale topic jobs before showing Task Center state.
Saving unrelated settings no longer clears an existing Profile when the save request did not include profile fields.
Library no longer crashes when opening the Wiki tab with the new topic refresh controls.
Library Wiki topic summaries are cleaner. Markdown is rendered in the topic panel, and graph-only fallback summaries no longer expose raw headings, list markers, or low-signal imported memory titles.
Wiki Topic Detection now defaults to AI-written summaries again. Structure-only refresh is still available as a cheaper mode, but it will not replace an existing AI-summarized Wiki when that would downgrade topic quality.
Knowledge Files now shows graph-only Wiki topics in the tree. Topic refreshes that skip AI-written summaries to save credits still appear as browsable topic folders instead of being hidden from the file view.
Knowledge Files time views now list real date buckets. Recorded At shows when memories entered Mem, and Happened On shows dates captured from memory event metadata.

0.9.11

2026년 6월 15일

0.9.112026년 6월 15일

More Targeted Search and Faster Live Thread Capture

Improved

0.9.10

2026년 6월 14일

0.9.102026년 6월 14일

Safer Backfills, Bulk Distillation, and Whole-Graph Export

0.9.10 makes large libraries easier to maintain: exports cover the whole graph, imported thread backlogs are planned in small safe waves, and background graph work pauses cleanly instead of timing out as a failure.

Added

Improved

0.9.9

2026년 6월 13일

0.9.92026년 6월 13일

Smarter Thread Reading, Clearer Background AI, and Memory Links

0.9.9 makes long-thread analysis more careful with what it reads, makes Processing history clearer when credits, quota, or provider credentials need attention, makes Thread search less likely to surface unrelated conversations, and makes user-created Memory Links visible in the graph overview.

Fixed

0.9.8

2026년 6월 13일

0.9.82026년 6월 13일

Faster Search, Clearer Background AI, and AI Now Minimap

0.9.8 keeps Mem responsive when managed AI or remote embeddings stumble, makes background task spending follow your Processing settings, reduces idle memory after heavy runs, and adds a minimap for long AI Now conversations.

Highlights

Added

Long AI Now conversations now include a minimap, so you can jump through message history without scrolling line by line.

Improved

Fixed

0.9.7

2026년 6월 13일

0.9.72026년 6월 13일

Startup, Background AI, and Search Hotfixes

0.9.7 is a hotfix for the 0.9.6 line. It fixes macOS startup hangs caused by damaged bundled Python runtimes, prevents valid ChatGPT Subscription GPT-5 models from being treated as tiny-context models, makes Memory Search steadier on specific questions, and removes a few sharp edges in Skills, Feed Timeline, and link-based skill imports.

Highlights

Improved

Fixed

0.9.6

2026년 6월 12일

0.9.62026년 6월 12일

Skill Imports and AI History Imports

0.9.6 makes it easier to bring more of your AI work into Mem. You can import multi-file skills from links or local folders, bring in Cursor Agent and Gemini/AI Mode history, and get a steadier AI Now, Background AI, provider, and Memory Search experience.

Highlights

Added

Fixed

0.9.5

2026년 6월 12일

0.9.52026년 6월 12일

Safer Background AI and Large Imports

0.9.5 tightens the reliability work behind 0.9: background AI usage is easier to understand, cost limits are harder to overrun, Memory search is more willing to stay quiet when evidence is missing, and large imports are safer on smaller machines.

Highlights

Improved

Fixed

0.9.4

2026년 6월 11일

0.9.42026년 6월 11일

Search, Skills, and Import Reliability

0.9.4 tightens the 0.9 release line: Memory search is now measured with a real recall-and-precision harness, imported skills keep their supporting files and tests, slow embedding paths no longer stall normal search, and several connector/import issues are fixed.

Highlights

Improved

Fixed

0.9.3

2026년 6월 11일

0.9.32026년 6월 11일

Background Work, Rules, and Connector Backfill

0.9.3 is a reliability patch for the 0.9 release line. Manual Background Intelligence runs no longer wait behind quiet maintenance work, Rules Review is more honest about generated drafts, coding-agent connectors gain preview-first history backfill, and local/self-hosted model setups are easier to tune.

Highlights

Added

Improved

Fixed

0.9.2

2026년 6월 10일

0.9.22026년 6월 10일

Thread Import and Pi Thread Sync Hotfix

0.9.2 is a hotfix for conversation history workflows after 0.9.1. It fixes large thread imports, Search Refresh on upgraded conversation stores, remote Feed actions, and Pi thread sync.

Highlights

Fixed

Fixed batch conversation imports from API and CLI paths so large imports stay searchable and available for background analysis without checkpoint storms.
Fixed Search Refresh for mixed conversation stores after upgrade, including cases where legacy rows are gone because messages were already safely copied into the Content Store.
Fixed Feed review and resolution actions in remote mode by routing them through the same transport map used by local clients.
Fixed the Pi connector path so Pi can sync conversation threads, not just use memory search and save flows.
Fixed a few small UI rough edges: thread message action icons are clearer, and Context primary buttons stay readable.

0.9.1

2026년 6월 10일

0.9.12026년 6월 10일

Background Tasks and Plus Login Hotfix

0.9.1 fixes two release-day rough edges from the 0.9 launch: Background Task Manager visibility and Mem Plus account sign-in. Settings now has a clearer view of recent background work, and Plus users get a more reliable path back into their account when a sign-in link or local session is no longer valid.

Highlights

Fixed

0.9.0

2026년 6월 9일

0.9.02026년 6월 9일

Nowledge Link, Mem Plus, Skills, and Context

0.9.0 is about three things: making Mem easier to run anywhere, teaching your AI how you work, and giving long-running agents stable identities and always-on rules. Nowledge Mem Plus adds included Nowledge AI credit and Nowledge Link, giving subscribers managed AI and a stable remote-access link while Mem data stays local. Skills turn repeated workflows into reusable procedures, while Context, AI Profiles, and Rules help each connected agent start with the right role and behavior before retrieval begins.

Nowledge Link shown inside the Nowledge Mem Plus account panel — Nowledge Link gives Plus users a stable remote-access address without manual tunnel setup.

Skills view in Nowledge Mem showing reusable working methods — Skills turn repeated workflows into reusable methods your AI can follow.

Context view in Nowledge Mem showing AI identities — Context gives long-running agents stable identities and always-on rules before retrieval begins.

Highlights

Added

Improved

Settings > Plans, Providers, and Access Anywhere now keep subscription status, Nowledge AI balance, and included services visible while account data refreshes in the background.
Nowledge AI account credentials self-heal across AI Now, background intelligence, provider checks, and managed model calls when an old account-issued key expires.
AI Now can now choose a different configured provider or model per task. The change applies on the next turn, keeps the task history and working folder, and does not rewrite your Settings default.
LM Studio, Xiaomi Mimo, Poe, Cerebras, Fireworks AI, NVIDIA NIM, and Jina AI are now part of the provider catalog, with browser Distill and Auto-Capture following the same routing as the desktop app where chat models are supported.
Background Intelligence now reviews new memories one by one, shows duplicate-review coverage, and treats label variants like knowledge-graph and knowledge_graph as the same idea.
Your skills stay current. When the way you work changes, Mem notices that a skill is now based on outdated steps and offers to update it with what is new, so a skill never quietly keeps teaching an old method.
The Skills home now tells you when Mem last looked for new skills and links straight to the automation settings, so the work happening in the background is visible and in your hands.
New to Skills? Mem can learn your first skills from the AI conversations you have already had. Bring in your existing chats and Mem starts spotting reusable procedures in them, so you are not starting from an empty page.
Skill suggestions are sharper. Instead of only saying a skill was learned from your work, each suggestion now names the specific thing your AI tends to get wrong without it, so you can tell at a glance whether it is worth turning on.
Agent-created memories now run through entity extraction, so background-created knowledge appears in graph views and entity search instead of staying isolated.
Settings now shows message-store size separately from the knowledge graph and search index, making it clearer whether disk usage comes from saved conversations or rebuildable search data.
Library PDF vision reading is more selective: diagrams, charts, screenshots, and tables can be enriched without rerunning full-page OCR on text-native papers.
Deep search now falls back to direct text and meaning matches when slow AI refinement would otherwise time out the whole search.

Fixed

AI Now approval popups no longer turn into accidental denials when you click outside the dialog or enable Auto while a tool is waiting.
AI Now starts correctly again on the current Kimi runtime and stays compatible with newer OpenAI-compatible provider behavior.
Auto-imported Codex, Claude Code, and OpenCode conversations now enter background analysis after the watcher saves them, so imported threads can become memories again.
Timeline is quieter and clearer: conversation syncs collapse into one completed analysis event, review actions preserve history, and selected cards carry their own evidence into follow-up turns.
Search repair is more reliable across macOS, Windows, and Linux, and now asks for a full rebuild only when embeddings or searchable content actually changed.
Search Optimize now prunes old native LanceDB full-text index versions after safe rebuilds, reducing the chance that repeated refreshes leave gigabytes of stale search files behind.
Daily briefings and scheduled background jobs retry cleanly after interruptions and stay quiet when a scheduled job is disabled.

0.8.8

2026년 5월 20일

0.8.82026년 5월 20일

Search Rebuild and AI Now Hotfix

0.8.8 makes full search rebuilds safer for users who changed embedding models or use OpenAI-compatible providers with smaller batch limits. It also fixes AI Now startup with strict Gemini/Google-compatible providers.

Highlights

Fixed

0.8.7

2026년 5월 20일

0.8.72026년 5월 20일

Search Refresh Hotfix

0.8.7 fixes a Search Refresh issue that could leave some upgraded users waiting at the final repair step when they had large Library documents. Conversation-search migration now stays lightweight and finishes without reparsing Library files.

Highlights

Search Refresh now completes the conversation-storage upgrade without getting pulled into large Library source repair.

Fixed

Fixed a recovery path where a lightweight conversation-search refresh could start reparsing large Library files, making the final repair stage appear stuck for a long time.

0.8.6

2026년 5월 15일

0.8.62026년 5월 15일

Knowledge Tree, Nowledge Mem FS, and Safer Search Refresh

0.8.6 introduces the preview of Knowledge Tree and Nowledge FS, and makes the 0.8.4 conversation-storage upgrade safer. Search Refresh now avoids unnecessary embedding rebuilds when only conversation search needs migration, recovers cleanly from interrupted 0.8.4/0.8.5 refreshes, and stays quiet on fresh installs with no content yet.

Knowledge Tree in Nowledge Mem showing memories, threads, wiki, working memory, activities, sources, and artifacts — Knowledge Tree gives every Mem object one familiar place to browse, with native previews instead of raw files.

Highlights

Added

Improved

Fixed

0.8.5

2026년 5월 15일

0.8.52026년 5월 15일

Search Refresh Hotfix

0.8.5 focuses on upgrade safety after 0.8.4. Search Refresh is less likely to run out of graph memory, completed refreshes no longer look failed when old conversation-storage cleanup has to wait, and stale refresh prompts from 0.8.4 are cleared when Mem can verify the rebuild already finished.

Fixed

0.8.4

2026년 5월 14일

0.8.42026년 5월 14일

Docker Preview and Leaner Conversation Storage

0.8.4 is mostly a refactor and reliability release. Mem moves thread messages out of Knowledge Graph storage and onto a lighter Content Store path, introduces the official Docker deployment preview with web-triggered operator-sidecar updates, and fixes a long list of Linux headless and remote-access edge cases.

Highlights

Added

Improved

Fixed

0.8.3

2026년 5월 2일

0.8.32026년 5월 2일

Plugin Refactor and DeepSeek V4

0.8.3 tightens the coding-agent plugin layer and adds DeepSeek V4 support. Plugin capture now follows one clearer contract across major hosts, Codex can save real sessions, Claude Code and Cursor keep repeated captures on the same thread, and DeepSeek V4 works as a first-class model choice with safer tool-loop handling.

Highlights

Added

Improved

Fixed

0.8.2

2026년 5월 1일

0.8.22026년 5월 1일

Intel Mac Search Index Hotfix

0.8.2 is only for Intel macOS desktop builds. It fixes a search-index startup and rebuild failure caused by the older Intel-compatible LanceDB package. Apple Silicon macOS, Windows, Linux, mobile, CLI, and server packages stay on 0.8.1.

Highlights

Intel Mac users can initialize and rebuild the search index again. Mem now supports the older LanceDB table-listing API shipped in the Intel-compatible runtime.
No other platform needs this hotfix. If you use Apple Silicon macOS, Windows, Linux, mobile, CLI, or a standalone server package, stay on 0.8.1.

Fixed

Intel macOS builds no longer fail search indexing with `LanceDBConnection object has no attribute list_tables`. Mem now works with both the current LanceDB `list_tables()` API and the older `table_names()` API used by the Intel-compatible package.

0.8.1

2026년 5월 1일

0.8.12026년 5월 1일

ChatGPT MCP, Web AI Now, and Remote Fixes

0.8.1 focuses on remote-first work: ChatGPT Web and desktop can connect to Mem over Access Anywhere with OAuth, AI Now sessions live on the Mem server for desktop, web, and mobile clients, browser folder import works in Library, and several Access Anywhere, updater, Feed, and CLI edge cases are fixed.

Read the release blog post

Nowledge Mem settings showing the ChatGPT Web OAuth connector setup — Start the ChatGPT Web connector from Mem, then approve access through OAuth instead of pasting a private API key.

ChatGPT using the Nowledge Mem connector from ChatGPT Web — After approval, ChatGPT can call Mem over Access Anywhere with `@Nowledge Mem`.

Highlights

Added

ChatGPT and other OAuth-capable MCP clients can now connect to Mem through Access Anywhere without a private `nmem_` API key. For ChatGPT, create the connector from web settings, then use it in ChatGPT Web or desktop with `@Nowledge Mem`.

Improved

Fixed

0.8.0

2026년 4월 29일

0.8.02026년 4월 29일

Mem Becomes Your LLM Wiki

0.8.0 brings the LLM Wiki pattern fully inside Mem. Topic, entity, and crystal pages read like real wiki entries; `[[Entity Name]]` wikilinks are clickable everywhere; Study with AI turns a source into saved memories or crystals you decide to keep; Wiki Export ships your knowledge as portable markdown. Investigate carries any wiki page into the Knowledge Graph with Graph Intelligence already aimed at the right cluster. This release also adds scanned-PDF reading and closes reliability gaps in Spaces, Memories, AI Now recall, search rebuilds, provider routing, and Windows startup.

Read the release blog post

The Library became a wiki: topic clusters, entity pages, crystals, and graph investigation in one reading path.

Highlights

Added

Improved

Fixed

0.7.10

2026년 4월 23일

0.7.102026년 4월 23일

Lower Rebuild RAM and Safer Linux Updates

0.7.10 focuses on the heavy maintenance paths. Search rebuild uses less memory, update downloads are faster, and Linux installs recover more clearly when the runtime, graphics stack, or CPU is the real blocker.

Highlights

Improved

Fixed

Windows search-model cleanup now removes old pre-0.7.9 cache duplicates once the new active model is healthy. Upgraded Windows installs should no longer keep paying disk space for both cache locations.

Platform

Linux package installs now switch backend code forward together with the desktop app. After a DEB or RPM update, Mem prefers the healthy packaged Python runtime instead of accidentally staying on an older fallback runtime left under `~/.local/share/nowledge-mem/runtime`.
Some Linux desktops no longer need manual `.desktop` edits just to get past a first-launch white screen. Mem now auto-applies the safer WebKit startup path not only for Arch-like systems, but also when it detects fragile renderer stacks such as NVIDIA-backed Linux or virtualized desktops like VMware.
Older Linux `x86_64` machines now fail clearly instead of crashing with a bare `Illegal instruction`. The current desktop package requires an AVX2-capable CPU, and Mem now says that directly before startup instead of dying opaquely on older non-AVX2 Xeon / Ivy Bridge-class machines.

0.7.9

2026년 4월 22일

0.7.92026년 4월 22일

AI Now Runs Smoother, Search Rebuilds Stay Stable

0.7.9 removes friction from two heavy paths: long AI Now research runs and large search rebuilds. Research stays readable and responsive, rebuilds are less likely to spike RAM or lose their place on smaller machines, and the browser extension also adds Coze and Yuanbao capture.

Highlights

Added

Nowledge Mem Exchange now supports both Coze and Yuanbao. Captured chats are recognized as first-class sources, with cleaner message extraction and the right source identity inside the app.

Improved

Fixed

0.7.8

2026년 4월 21일

0.7.82026년 4월 21일

A Better Graph Overview, Cleaner Labels, and Easier Copilot Setup

0.7.8 makes the Graph easier to navigate. You now open into a calm overview of your whole knowledge, uploaded documents are called Artifacts throughout the graph, and Copilot CLI becomes a proper built-in integration.

Added

Fixed

0.7.7

2026년 4월 20일

0.7.72026년 4월 20일

Library Search Reaches More Places

0.7.7 brings Library search to more AI surfaces and tightens a broad set of reliability issues. AI Now, Feed, Graph Intelligence, and `nmem` can all reach into your documents, while search recovery, updater behavior, and provider diagnostics all get less fragile.

Highlights

Added

Fixed

0.7.6

2026년 4월 17일

0.7.62026년 4월 17일

Search Recovery and IME Input Hotfix

0.7.6 fixes two problems that show up fast in real use: search rebuild recovery on upgraded indexes, and accidental message sends while typing with CJK IME input.

Highlights

Fixed

0.7.5

2026년 4월 16일

0.7.52026년 4월 16일

Bulk Space Cleanup and Export Reliability

0.7.5 makes large space cleanups more practical. You can move or clean up Memories and Threads in bulk, export is steadier on busy setups, and OpenAI-compatible providers plus localhost browser-extension access both behave more predictably.

Highlights

Added

Fixed

0.7.4

2026년 4월 14일

0.7.42026년 4월 14일

Search, Thread Sync, and Remote Clients Feel More Trustworthy

0.7.4 is a reliability release. Search explanations stay more honest, thread sync causes less UI drag, and desktop, web, remote, and mobile clients stay closer to the same state.

Highlights

Improved

Fixed

Platform

0.7.3

2026년 4월 12일

0.7.32026년 4월 12일

Follow-up Hotfix

0.7.2

2026년 4월 12일

0.7.22026년 4월 12일

Spaces: Timeline & Feed Hotfix

0.7.2 fixes a regression from Spaces where Timeline and the AI Feed could use the wrong lane or show empty previews. Your feed and timeline again follow the active space consistently with the rest of Mem.

Highlights

Fixed

Fixed a Spaces-related regression where Timeline and AI Feed queries could ignore or mis-bind the current space, causing wrong or empty results until this hotfix.

0.7.0

2026년 4월 12일

0.7.02026년 4월 12일

Introducing Spaces

0.7.0 introduces Spaces, the first real way to separate projects, purposes, and agent tendencies inside Mem without forcing everything into one shared pool. The same release also brings a broad performance pass, so larger knowledge bases feel faster.

Highlights

Added

Improved

Fixed

Space-aware behavior now stays more consistent across desktop, web, remote, and CLI paths, reducing lane drift between client modes.

0.6.19

2026년 4월 8일

0.6.192026년 4월 8일

Graph Reasoning, Simpler Remote Setup, and Better Recovery

0.6.19 turns the graph into a place to work, not just look. The new Reason panel lets an AI agent explore your graph step by step on canvas, while remote setup, Linux headless deployment, Windows updates, and CLI text handling all get sturdier.

The Knowledge Graph Agent turns graph view into a reasoning workspace, with each step replayable on the canvas.

Highlights

Added

Improved

Fixed

Platform

0.6.18

2026년 4월 7일

0.6.182026년 4월 7일

A Faster Graph, Vision in AI Now, and Plugin Fixes

0.6.18 makes the graph much easier to explore, lets AI Now accept pasted images, and cleans up reliability across Alma, OpenClaw, Hermes, Codex, and remote settings sync.

Highlights

Added

Fixed

Alma plugin no longer freezes the host app. All data operations use async HTTP instead of blocking CLI calls that hung the event loop for seconds. Also fixes thread duplication on restart and broken search/recall/dedup from parameter mismatches.
When you use Mem over a remote connection, Token Budget and other background-intelligence settings now stay in lockstep with your server: what you see is what runs, and changes you save actually apply on the machine hosting your knowledge base.
Browser-extension and imported ChatGPT threads now share the same identity path, so extension-synced and imported versions of the same conversation no longer split into duplicates
Graph importance no longer drops to 0% after PageRank recalculation. Importance now always reflects the original knowledge importance; PageRank is used only for layout sizing.
Graph shortest path query works correctly with KuzuDB (previously failed due to parameter binding in path bounds)
Background intelligence tasks no longer get stuck in retry loops when the agent sends empty search queries
AI Now clipboard image paste survives real desktop sends and large screenshots no longer overflow the transport connection
Codex plugin: YAML skill descriptions no longer break serde_yaml parsing, and skill naming adjusted for Codex compatibility
Hermes plugin: fixed install path (was writing to wrong directory) and Python import errors under Hermes' importlib loader
OpenClaw plugin: Working Memory now injects on short messages (previously skipped on messages under 3 characters), and scheduled cron runs no longer crowd your Threads list
Feed and background tasks now respect your preferred output language setting. Daily briefings, insights, and crystals are generated in the language you chose in your profile.

0.6.17

2026년 4월 5일

0.6.172026년 4월 5일

Hermes Abstraction, Safer Upsert, and Better Codex Setup

0.6.17 is mainly reliability work. Hermes gets a cleaner memory-plugin path, Codex installs behave better, and upsert-by-ID is safer across CLI and integration workflows.

Highlights

Added

Improved

Fixed

0.6.16

2026년 4월 3일

0.6.162026년 4월 3일

OpenCode Support and Lower RAM Pressure

0.6.16 adds the OpenCode plugin and the new Exchange browser extension release, while cutting idle RAM use and making search and database memory behavior much steadier.

Highlights

Added

Improved

Fixed

0.6.15

2026년 4월 1일

0.6.152026년 4월 1일

Remote Embeddings and a Friendlier CLI

0.6.15 expands remote-first workflows. Search indexing can use remote embedding providers, the CLI becomes easier to automate, and AI Now, graph, search, and background jobs all get a broad stability pass.

Highlights

Added

Improved

Fixed

0.6.14

2026년 3월 31일

0.6.142026년 3월 31일

Remote Embeddings, Agent-Friendly CLI & Reliability Upgrade

This release expands remote-first workflows and tightens stability across AI Now, graph, search, and background processing. You can now run search indexing with remote embedding providers (Pro), use a much more automation-friendly `nmem` CLI, and rely on calmer AI Now sessions with stronger retrieval and rendering behavior.

Highlights

Added

Fixed

0.6.13

2026년 3월 24일

0.6.132026년 3월 24일

Auto-Learn and a Better Context Picker

0.6.13 makes Mem quicker to use day to day. Library items can learn automatically after indexing, you can pull existing memories, sources, or threads into a prompt more directly, and Claude Desktop reconnects with less friction.

Highlights

Turn on Auto-learn in Settings → Memory Processing, and new sources extract knowledge automatically after indexing — no more clicking Learn one by one
Attach sources, memories, or past conversations when composing in the Feed — search across all your knowledge, pick what's relevant, and ask questions grounded in context
AI Now now reads your own history more naturally: broad prompts can pull from recent threads and show cleaner thread/library cards instead of raw tool JSON
AI Now sessions feel calmer under heavy tool output, with less redraw noise and fewer unnecessary follow-up questions before first lookup
Claude Desktop extension now installs on macOS and Windows with no separate Python or Node setup
OpenClaw long-running sessions now sync reliably — large repetitive thread backups no longer fail on oversized CLI payloads
Mobile and web apps detect when your server goes offline and show a clear status screen instead of a technical error message
On macOS, you can now keep Mem in the menu bar after closing the main window; the Dock icon returns automatically when you open the app again
Labels now propagate to crystals, merged knowledge, and search results — label-based filtering works reliably everywhere
Browser extension capture is now much smarter: Manual Distill can create or update multiple memories from one conversation, and browser-created memories now keep a reliable link back to the full thread

Added

Auto-learn sources: an opt-in toggle in Settings → Memory Processing that automatically extracts knowledge from new sources after indexing. Manual Learn always takes priority over auto-learn tasks
Context picker in the Feed: a new button in the Feed toolbar lets you search and attach Library sources, memories, or past conversations when composing. Selected items appear as chips and the resulting memory links back to them
Labels on ingestion: the upload API, file-path ingestion, and batch endpoints now accept labels — tag sources at the point of capture from the CLI, integrations, or the API
Claude Desktop extension now installs on macOS and Windows with no separate Python or Node setup. For remote access, it reads the same shared config file as the `nmem` CLI
New macOS Menu Bar & Dock preference: after you close the main window, Mem can stay in the top menu bar instead of remaining in the Dock. When you reopen the window, the Dock icon comes back automatically
The `nmem` CLI is now automatically configured when the desktop app generates an API key — no more manually copying credentials
Connect your phone in seconds: the desktop app can now show a QR code in Access Anywhere settings. Scan it from the mobile or web login screen to connect instantly — no more typing long URLs and API keys
`nmem status` and the TUI now show whether you're connected locally or to a remote server — no more guessing which Mem you're talking to
Sources in the Library can now be organized with labels — the same label system used for memories now extends to sources, so you can tag, filter, and group your reference material

Fixed

AI Now now uses recent conversation threads and correct Library source-reading tools more reliably, so broad history questions don’t depend only on Working Memory
Thread and Library retrieval results now render as calmer dedicated cards instead of raw JSON-like output
Memory, graph, community, crystal, Working Memory, and thread cards now use more consistent spacing and headers across AI Now
Memory mode and plan/graph card routing are more stable, with fewer scroll-position jumps during graph-heavy conversations
Reduced accidental hot-path logging and unnecessary graph redraws make long tool-heavy AI Now sessions less flickery
Obsidian vault lookup is less likely to stall: direct vault-name candidates are tried before full vault listing
Browser extension Manual Distill now reads conversations more intelligently instead of flattening them into one quick save — it can inspect the thread selectively, check what already exists in Mem, and create or update multiple memories when that leads to a better result
Browser-created memories now keep stronger provenance — when the extension writes a new memory, it first makes sure the conversation is backed up as a canonical thread in Mem, so you can trace where that knowledge came from later
Labels now carry through to crystals and merged knowledge — when Background Intelligence distills or merges memories, the labels from source memories come along. Label-based filtering is now reliable across all knowledge types
Memory search results now include labels — integrations like Claude Code and Cursor can see and filter by labels in search results
Mobile and web clients no longer show a raw "Load failed" error when the server is unreachable — a clear reconnection screen appears instead, with automatic retry every few seconds
If the server goes down while you're using the app, you're taken to the reconnection screen rather than seeing broken views — the app resumes right where you left off once the server is back
Desktop update checks are now platform-aware. If a new build is available for macOS first, Windows and Linux devices keep seeing their own newest available build instead of a version they cannot install yet
Thread source filters now work across your whole conversation history. The source menu lists every imported source, and selecting one refreshes the full paginated thread list instead of filtering only the current page
The Threads source menu now scrolls cleanly when you have many sources, and icons like Gemini CLI, OpenClaw, Antigravity, Grok, Manus, and OpenWebUI render correctly
Custom providers (SiliconFlow, together.ai, and other OpenAI-compatible services) now work reliably in AI Now — previously they could pass the connection test in Settings but fail when you started a conversation
AI Now no longer hangs on startup when the agent process fails to launch — the error now appears immediately with diagnostic details instead of an infinite loading spinner
GitHub Copilot with Codex models now works in AI Now — these models are automatically routed to the correct API endpoint
GitHub Copilot headers updated across all integrations to match the current API, fixing authentication errors that appeared with newer Copilot releases
ChatGPT Subscription models now work for background knowledge extraction — previously the request was silently misrouted and fell back to local processing
OpenRouter models no longer fail with 400 errors during knowledge extraction — the provider is now correctly identified so requests go to the right API endpoint
The desktop app no longer gets stuck on a white screen during startup when the loading view hands off to the main app
The `nmem` CLI no longer shows a cryptic "Resource doesn't exist" error when the server is unreachable — it now explains the problem clearly and, for first-time remote users, shows exactly how to set up `~/.nowledge-mem/config.json`
Server URLs with a trailing slash (e.g. `https://mem.example.com/`) no longer cause silent connection failures across the CLI, TUI, and all integrations
OpenClaw long and repetitive sessions now back up reliably — thread sync no longer tries to squeeze whole conversation histories through one oversized CLI command, so repeated cron-style captures keep syncing instead of failing mid-session

0.6.12

2026년 3월 17일

0.6.122026년 3월 17일

Crash-Loop Recovery, VPN Connectivity & Mobile Fixes

Fixes crash-loop recovery, improves remote connectivity for Tailscale and VPN users, and resolves Android mobile app installation issues.

Highlights

Added

Fixed

0.6.11

2026년 3월 16일

0.6.112026년 3월 16일

iPhone Remote Access and Smarter Remote AI

0.6.11 brings iPhone access to a remote Mem server, lets AI Now use the provider already configured on that server, and fixes a wide set of reliability issues across CLI auth, MCP transport, and provider behavior.

Highlights

Added

Improved

Fixed

0.6.10

2026년 3월 13일

0.6.102026년 3월 13일

Windows In-App Updates and Better Remote Access

0.6.10 brings Windows in-app updates to parity with macOS and tightens remote access across web and CLI flows. It also fixes the rough edges around mobile-browser remote access.

Highlights

Added

Server capabilities endpoint lets remote clients adapt their UI based on what the server supports

Improved

Fixed

0.6.9

2026년 3월 12일

0.6.92026년 3월 12일

Remote Mem from Desktop, Web, and Raycast

0.6.9 makes remote access feel like a real product surface. The desktop app can connect to a remote Mem as a client, Raycast can reach your Mem from anywhere, and the Graph API plus Windows startup behavior both get sturdier.

Highlights

Added

Improved

Fixed

Platform

Agents like Claude Code running inside WSL can now use `nmem` directly — the Windows CLI installer creates a lightweight bridge script that connects to the desktop app without additional network setup

0.6.8

2026년 3월 10일

0.6.82026년 3월 10일

Lower Memory Footprint, Smaller Indexes & Startup Hardening

This release dramatically cuts memory and disk pressure for long-running desktop usage, while fixing reliability issues in deep-learn, background task retry, Access Anywhere readiness, and Windows startup flows.

Highlights

Added

Improved

Fixed

0.6.7

2026년 3월 10일

0.6.72026년 3월 10일

Sharper Integrations and Safer Access Anywhere

0.6.7 makes setup less brittle across Gemini CLI, Cursor, browser capture, and remote thread save paths, while improving everyday diagnostics and import reliability.

Highlights

Added

Improved

Integration pages now use clearer product language, calmer visual hierarchy, dedicated command blocks, and more useful first-time setup guidance
Search ranking now accounts for memory confidence from real usage signals like search, open, and read, helping well-validated memories surface earlier
Automatic labeling now applies on creation and reuses your existing label vocabulary for more consistent organization
AI Now `@` picker now spans memories, threads, and Library sources in one place
Large file workflows in AI Now are faster: the agent now prefers search and sample-first analysis before full reads
Library status language is clearer: "Searchable" versus "Deep-learned" with filter support and batch deep-learn actions

Fixed

Thread save reliability improved: `nmem t save --from claude-code|codex|gemini-cli` now parses local session files without requiring server-side agent-path access
Gemini CLI now separates full thread save from resumable handoff summaries, making capture behavior clearer during long coding sessions
Subscription status and premium gating now use one unified access contract across Plans, AI Now, Remote LLM, and backend checks, reducing misleading states
Windows second-launch boot failure fixed for 0.6.5 and 0.6.6 upgrade paths, including stale CLI wrapper and stale-port recovery handling
AI Now now surfaces provider and startup failures inline and keeps chat context visible, with copyable diagnostics instead of silent hangs
Search results now show correct confidence values, separate from relevance score, and agent-assigned labels are visible consistently across surfaces
Plugin toggles in AI Now now actually filter tools, including custom MCP tool dedup fixes
ChatGPT Subscription and GitHub Copilot OAuth flows now work correctly in the CLI and TUI test-connection flow
Memory distillation no longer fails to blank output silently: it falls back to the local LLM or returns clear failure details
Browser extension reliability updates: Grok detection on x.com and improved unsupported-browser messaging for failed AI-access checks

Platform

Linux package uninstall parity: both `nmem` and `browse-now` symlinks are now cleaned up consistently
Startup path resolution is more resilient across macOS, Windows, and Linux after updates, including plugin, CLI, and Python fallback paths

0.6.6

2026년 3월 4일

0.6.62026년 3월 4일

Windows Startup Hotfix and Provider Reliability

0.6.6 removes a Windows startup blocker, improves Intel Mac compatibility, and makes AI Now plus remote-model workflows less fragile when providers misbehave.

Highlights

New experimental Data Transfer: export a portable snapshot and import it later with merge/skip/overwrite modes and automatic backup
Critical Windows PATH corruption fix: startup no longer rewrites expandable PATH entries and breaks other CLI tools
Windows idle GPU burn is eliminated by making the aurora background static
Startup reliability is stronger across Linux/macOS/Windows, including RPM runtime self-healing and safer DB path recovery
Obsidian integration now uses official CLI workflow with robust fallbacks for real-world vault edge cases
Claude Code plugin v0.7.0: full lifecycle hooks for Working Memory, session capture, and per-turn search/save nudge

Added

Data Transfer (Experimental): export a portable snapshot and import into another install with merge/skip/overwrite options, selective inclusion toggles, and a safety backup before import
Claude Code plugin v0.7.0: Working Memory loaded at session start/resume/clear, async session capture after every response, and per-turn behavioral nudge with search/save syntax
nmem-cli config file support: set remote connection once in ~/.nowledge-mem/config.json (flags > env > config > defaults). `nmem status` now shows where values came from.

Improved

Windows GPU usage is now near-zero at idle: aurora background is static by default, removing continuous recomposition load
AI Now and secondary surfaces now use a more consistent Phosphor icon system with unified sizing/weight behavior
v0.6 UI/ICON polish across thread detail cards, glowing cards, and pagination for cleaner visual hierarchy
Obsidian setup guide now uses full-screen screenshot preview for faster first-time setup
Obsidian plugin now uses official CLI tools as source of truth for better multi-vault consistency

Fixed

Platform

0.6.5

2026년 2월 28일

0.6.52026년 2월 28일

Windows Startup Recovery & Intel Mac Reliability

This hotfix removes a startup blocker on some Windows machines, improves Intel macOS compatibility, and tightens provider reliability so AI Now and remote-model workflows fail less often and recover faster.

Highlights

Known issue (0.6.5): AI Now startup can fail on macOS, Windows, and Linux; dedicated hotfix is in progress
Windows users now get a direct in-app fix path when Microsoft VC++ runtime is missing
Intel macOS 15+ startup checks and search embedding path now work consistently
MiniMax Anthropic endpoint defaults are safer, reducing AI Now 404/provider mismatch failures
Configured providers can now be removed safely in both Desktop Settings and `nmem` CLI/TUI

Fixed

Windows startup no longer dead-ends on missing VC++ runtime (`_lbug` DLL load): Startup Flow now shows clear remediation links and the same guidance is mirrored in logs for support teams
Intel Macs on macOS 15+ are no longer incorrectly blocked before backend initialization
Intel macOS search embeddings now follow the correct non-Apple-Silicon path end-to-end, improving model download, verification, and reindex stability
Local LLM capability detection on macOS is now architecture-aware: Intel Macs correctly guide users to Remote LLM workflows
MiniMax provider defaults now favor Anthropic-compatible endpoints and remove incompatible presets that caused cross-surface failures
Moonshot CN K2.5 extension tool-loop requests now use valid payload semantics, fixing 400 errors in capture workflows
Token budget settings via `nmem config settings set` now take effect immediately — previously the API silently dropped these values so only GUI changes worked

Improved

Provider lifecycle is cleaner across GUI and terminal workflows: delete providers safely and let purpose fallback resolve automatically
Intel macOS search model metadata is now shown with correct platform/backend expectations

Platform

Known issue: AI Now startup regression can prevent AI Now session boot on all platforms in 0.6.5; dedicated hotfix is in progress
Bundled Ladybug fork is rebased to upstream 0.15.0 while preserving desktop-specific integration behavior

0.6.4

2026년 2월 28일

0.6.42026년 2월 28일

Smarter Tables and Faster Long Lists

0.6.4 turns CSV and XLSX files into something you can actually work with: charts, cleaner table rendering, and better searchable chunks. It also makes large libraries and long timelines easier to browse, with more provider control from the CLI.

Highlights

CSV/XLSX now auto-generate interactive charts in Feed and Library
Timeline now loads in larger batches and can jump to any calendar date on demand
Library source list now paginates progressively, keeping large collections responsive
Agent provider/model purpose can now be configured fully from `nmem` CLI/TUI
DeepSeek is now supported across desktop settings, AI Now, CLI/TUI, and extension flows
Linux `.deb` installs now support signed APT updates and safer upgrade shutdown
Apple Silicon idle memory drops significantly with automatic model unload when inactive
OpenAI-compatible local Embeddings API for tools and scripts

Added

OpenAI-compatible local Embeddings API (`/v1/embeddings`, `/v1/models`) for external tools and automation
Agent source management: Feed and Background Intelligence can now discover and delete Library sources safely in one workflow
New `nmem sources` CLI commands (`list`, `show`, `delete`) for source lifecycle control from terminal workflows
Terminal-only provider purpose control: `nmem` CLI/TUI can now set default provider and Agents (`ai_now`) purpose provider/model without opening GUI Settings
DeepSeek provider support is now end-to-end across app settings, `nmem` CLI/TUI, AI Now, and browser extension flows
Tabular intelligence: auto chart detection plus schema-based chunking makes CSV/XLSX content more searchable and answerable
Claude Desktop extension (v1.3.0): configure Remote URL + API Key directly in extension settings for Access Anywhere
Library pagination: source list loads progressively with "Load more", keeping large libraries fast
Timeline "Load more": feed loads 200 items initially and grows on demand, with calendar-aware date fetching for any historical day
Linux signed APT repository for `.deb` users: one-time setup enables normal `apt upgrade` updates
Safer Linux package upgrades: app now checkpoints and shuts down gracefully before package replacement to protect database state
AppImage auto-update now supports download, install, and restart directly in-app (same flow as macOS)

Improved

Search remains responsive during heavy Background Intelligence write bursts with per-table index rebuilds and debounce scheduling
Table reading experience upgraded across Feed, Library, and AI Now with sticky headers, better scrolling, and clearer numeric alignment
Feed Agent table analysis now reads and reasons in a single pass, producing clearer analytical answers and visualizations
Library Chat keeps QA history when switching files and across window reloads, with per-source isolation
Timeline calendar click now navigates to any date, fetching just that day's events on demand instead of requiring all items pre-loaded
Windows WebView graphics usage reduced by removing redundant animated layers and pausing animations when app is not visible
Apple Silicon memory behavior improved for long-running sessions: inactive local models auto-unload and reload when needed, reducing idle RAM

Fixed

Feed Agent context amnesia fixed: recent sessions persist (time/turn bounded) and timeline summaries are injected for continuity
Source deletion dead-ends resolved: agents can now find source IDs, handle parse-failed sources, and avoid follow-up loop explosions
Token budget false positives removed: budget checks now match real context usage and limits are configurable in Settings and CLI
Large file preview no longer freezes Library UI; previews are capped and rendered progressively
macOS updater no longer hard-kills the backend process, preventing Kuzu WAL corruption risk during install
macOS update confirmation now uses native Tauri dialog for reliable user consent before install
TUI provider selector no longer crashes on configured custom provider IDs due to Select markup parsing
Moonshot CN K2.5 tool-loop reliability improved in browser extension: force-capture now uses Moonshot-native thinking disable semantics instead of invalid reasoning-effort values
Vega-Lite charts no longer render at zero height in Feed and AI Now responses

Platform

0.6.3

2026년 2월 26일

0.6.32026년 2월 26일

Apple Notes, More Capture Sources, and Better Windows Startup

0.6.3 brings Apple Notes into AI Now, lets you connect multiple Obsidian vaults and Notion workspaces, adds display zoom, and makes Windows runtime setup more dependable.

Highlights

Apple Notes: search and read your notes directly from AI Now
Multi-instance: connect as many Obsidian vaults and Notion workspaces as you need
Display zoom: scale the entire interface to your preference
OpenClaw plugin: trace any memory back to the full conversation it came from
Alma plugin: status diagnostics and live settings — change credentials without restarting

Added

Apple Notes plugin for AI Now: search and read notes from macOS Notes.app — no export needed
Multi-instance Obsidian and Notion: connect multiple vaults or workspaces, each with independent config and auth
Display zoom in Preferences: Small, Default, Large, or Extra Large — scales the whole interface
Token budget controls in CLI and TUI: view and set hourly/daily limits from the terminal
OpenClaw plugin 0.6.5: thread search and progressive message retrieval across 9 tools — trace memories back to full conversations
Thread API pagination: fetch specific pages of long conversations instead of loading everything at once
Alma plugin 0.6.3: status diagnostics, live settings reload, and Access Anywhere support — 12 tools total

Improved

Notion plugin handles unexpected parameters gracefully and validates tool schemas more robustly
Plugin context uses uniform block tags — cleaner message rendering across all plugins
Agents model routing: OpenRouter tool-call validation and model ID normalization
npx Skills and Codex prompts updated with structured memory types and labels

Fixed

Windows Store: bundled Python runtime works reliably under MSIX packaging
Downloaded models found correctly on all platforms after cache path alignment
TUI: embedding model status no longer shows "Error - None"
Browser extension: Gemini folded messages and collapsed content captured correctly

Platform

Windows: CLI wrappers (nmem, browse-now) installed to user PATH for Store and non-Store builds
macOS: Apple Notes integration — read-only, iCloud-aware

0.6.2

2026년 2월 25일

0.6.22026년 2월 25일

Doubao, WSL2 & Smarter Capture

Sync Doubao conversations, discover WSL2 sessions on Windows, and enjoy dramatically lower background costs with intelligent capture controls.

Highlights

Doubao (豆包) conversations now sync through Exchange
Session Browser finds conversations inside WSL2 on Windows
OpenClaw autoCapture uses up to 95% fewer tokens when idle
Imported threads show clean tool summaries instead of raw output

Added

Doubao (豆包) support in Exchange: sync and export your conversations from doubao.com
WSL2 session discovery on Windows: Session Browser and Auto-Sync now find Claude Code, Codex, and OpenCode sessions inside WSL2 distributions
Multi-instance Obsidian & Notion plugins: connect multiple Obsidian vaults or Notion workspaces simultaneously, each with independent config and auth

Improved

Imported threads show concise tool summaries instead of raw grep results and file dumps
MCP Live Traffic in Integrations now displays real calls and responses

Fixed

OpenClaw plugin: autoCapture no longer burns tokens from idle heartbeat cycles (5-layer cost defense, ~95% reduction)
OpenCode sessions re-import correctly after deletion from Session Browser
MCP Live Traffic log parsing updated for current backend format

0.6.1

2026년 2월 24일

0.6.12026년 2월 24일

Cost Controls & 20+ Languages

See what background intelligence costs, set your own limits, and let the system pause before you overspend. Knowledge augmentation now works in 20+ languages.

Highlights

Token budget with hourly and daily limits, plus a live usage dashboard
Knowledge augmentation in 20+ languages including Chinese, Japanese, Korean, French, and German
Thread viewer handles 4,000+ messages without lag
macOS reindex uses up to 97% less GPU memory

Added

Token budget controls: set hourly and daily spending limits for background intelligence
Live token usage in Settings, the status API, and nmem status CLI
Knowledge augmentation in 20+ languages: insights, briefings, and extractions in the language you choose
Kimi Coding Plan provider support

Improved

Thread detail view handles 4,000+ messages smoothly
Reindex on Apple Silicon uses up to 97% less GPU memory
Smarter PDF detection: arXiv-style links and /pdf/ routes go through the file pipeline automatically
Per-task safety net prevents single background tasks from running away
Lower memory footprint during indexing on Windows and Linux

Fixed

Preferred output language now persists correctly across restarts
Browser Extension: Gemini table data captured correctly
Feed: rapid consecutive submissions no longer dropped
Feed: failed URL captures no longer create broken memories
Settings: correct local model name shown instead of unrelated models
Background Intelligence sessions recover gracefully after unexpected exits
Direct PDF URLs from timelines now ingest through the file pipeline

0.6.0

2026년 2월 18일

0.6.02026년 2월 18일

Background Intelligence, Smart Feed & Library

Your knowledge now lives in a neutral layer that works with any AI tool. Switch freely between Claude Code, Cursor, ChatGPT, or whatever ships next. Background Intelligence builds connections and writes your daily briefing. Smart Feed, Library, and Exchange v2 round out the release.

Read the release blog post

Highlights

Smart Feed: one input for captures, questions, and file imports
Background Intelligence: the system finds connections, surfaces contradictions, and writes daily briefings on its own
Working Memory: a daily briefing of what matters, delivered to ~/ai-now/memory.md
Library: search PDFs, Word docs, and presentations alongside your memories
Access Mem Anywhere: one-click secure remote URL + API key for your other machines and agents
Exchange v2 (Chrome): auto-captures AI conversations across 13+ platforms
Browse-Now (Chrome): give AI agents access to your browser
Search engine rewritten in Rust with new BM25, vector search, and SOTA local embeddings
Alma plugin: native memory integration with tools, command palette actions, and auto-capture controls

Added

Smart Feed: one input for captures, questions, and file imports
Unified timeline: captures, insights, flags, and daily briefings in one stream
Background Intelligence: the system finds connections, surfaces contradictions, and writes daily briefings
Working Memory: daily briefing at ~/ai-now/memory.md, readable by any MCP agent
Memory evolution: detects when memories update, enrich, confirm, or contradict each other
Library: search PDFs, DOCX, PPTX, and Markdown alongside memories
Access Mem Anywhere: start a secure Cloudflare tunnel from Settings (Quick link or Cloudflare account mode)
Remote API auth hardening: every remote request requires your API key, with key rotation support
Exchange v2: auto-captures AI conversations across 13+ platforms (Claude, ChatGPT, Gemini, and more)
Browse-Now CLI: give AI agents access to your browser with your login sessions
Interactive graph visualization in Claude Desktop and other MCP hosts
Chinese full-text search
AI-powered search reranking with human-readable explanations
Session auto-sync: real-time watching for Claude Code, Cursor, Codex, and OpenCode
Linux headless server deployment with systemd
Full CLI for server setup, license, LLM config, and knowledge settings
New providers: ChatGPT (Codex) subscription, MiniMax, Z.AI, MoonShot AI
MCP: Graph View in Memory Search for Claude Desktop and other MCP-UI hosts
OpenClaw plugin: native memory integration for the OpenClaw agent framework
Alma plugin: native memory integration with tools, command palette actions, and optional auto-capture
Raycast extension: search memories, save insights, read Working Memory
npx Skills: one-command memory integration for any agent (search, save, Working Memory, graph exploration)
Knowledge structure: Trace → Unit → Crystal three-layer model inspired by cognitive science
Community detection: automatic topic clustering across the knowledge graph
Node importance scoring: surfaces the most influential concepts in your graph
LLM-friendly docs: every page serves clean Markdown via Accept: text/markdown header, plus /llms-full.txt and /llms.mdx endpoints

Improved

Search engine rewritten: new tokenization, index engine, and SOTA local embedding models
MCP: 24 tools exposing the full knowledge surface (up from basic search/add/labels)
Visual refresh: calmer, cleaner design with content-first layouts
Search is 50% faster with native Chinese support
Graph timeline slider: filter your knowledge graph by time range
Real-time streaming: see agent thinking as it happens

Fixed

Graph visualization performance improved
OAuth connections silently failing on macOS
Search Index Model downloads failing on Windows
Session export showing wrong message count from Cursor
Access Mem Anywhere copy actions now survive desktop reload/restart without forcing key rotation
Cloudflare account token parsing now accepts full command formats (`service install`, `--token`, `--token=`)

Platform

Linux: Full headless deployment with nmem serve and nmem service
Linux: Interactive TUI control plane for headless server management
Community: native support in DeepChat and LobeHub

0.5.11

2026년 1월 8일

0.5.112026년 1월 8일

MCP Browse Mode & CLI Thread Import

Browse recent threads and memories via MCP, and import Claude Code or Codex sessions directly from the CLI.

Highlights

MCP: Browse threads and memories without a search query
CLI: Import Claude Code and Codex sessions with `nmem t save`
Re-import conversations without duplicates

Added

MCP list mode: browse recent threads and memories without a search query
MCP source filter: filter threads by source (e.g., 'claude-code')
CLI `nmem t save`:import Claude Code and Codex sessions as threads
Memory search now shows which thread a memory came from

Improved

Claude Code skills now use nmem CLI for faster, more reliable operations
Codex custom prompts updated to use nmem CLI
Re-importing conversations appends new messages instead of failing
Automatic deduplication when re-importing the same conversation

0.5.10

2026년 1월 4일

0.5.102026년 1월 4일

Standalone CLI & Batch Thread Deletion

Access your Nowledge Mem from any machine with the new standalone CLI package.

Highlights

Standalone CLI: pip install nmem-cli
Remote access to your Nowledge Mem server
Batch delete multiple threads at once

Added

Standalone CLI package: access Nowledge Mem from any machine
Batch thread deletion: select and delete multiple threads
MCP thread deletion tool for AI integrations

0.5.9

2025년 12월 31일

0.5.92025년 12월 31일

AI Now Command Execution Fix

Fixes an issue with command execution in AI Now.

Highlights

Fixed AI Now command execution

Fixed

AI Now command execution not working correctly

0.5.8

2025년 12월 30일

0.5.82025년 12월 30일

Internationalization & Infographic Visualization

Full English and Chinese language support. Infographic visualizations in AI Now.

Highlights

Full English & Chinese language support
Infographic visualizations in AI Now responses
Improved database reliability and performance

Added

Internationalization: full English and Chinese translations
Language switcher in Settings with automatic system detection
Infographic visualizations for research and presentation skills
Plugin auto-updates when AI Now starts
macOS compatibility check for Apple Silicon and macOS 15+

Improved

Better detection and display of AI reasoning/thinking content
Markdown formatting: fixed newlines and structured content
Database reliability and data integrity on shutdown
Faster and smoother UI updates
Search and pagination performance

Fixed

Pro user status showing incorrectly when offline
Label changes being overwritten
Various markdown rendering issues in AI responses
Folder and file operation issues
Data loss issues

0.5.7

2025년 12월 23일

0.5.72025년 12월 23일

AI Conversation Discovery

Find and import your conversations from Claude Code, Cursor, and Codex with a single click.

Highlights

One-click discovery of AI coding conversations
Import from Claude Code, Cursor, and Codex
Searchable model dropdown for 20+ models

Added

AI Conversation Discovery: automatically find conversations from Claude Code, Cursor, and Codex
Preview conversations before importing (project name, message count, last modified)
Batch import: select and import multiple conversations at once
Searchable model select for providers with many models

Improved

Model selection shows filtered/total count for easier navigation
Faster parsing of large conversation files

0.5.6

2025년 12월 23일

0.5.62025년 12월 23일

16x Larger Memories & AI Now Stability

16x larger memory capacity, better AI Now stability with integrations, smoother UI.

Highlights

16x larger memory capacity (up to 32,768 characters)
Improved AI Now stability with integrations
Smoother UI with stable pagination

Added

16x larger memories: up to 32,768 characters per memory
GLM-4.7 and Gemini 2.5 Flash model support

Improved

AI Now stability with Notion and other integrations
Better search for long memories
Tool compatibility with Groq, OpenRouter, and more providers
More reliable AI responses across providers
Presentation and Travel Planner skill instructions
Smoother pagination and loading states

Fixed

Notion integration errors (400 Bad Request)
Remote LLM toggle not taking effect

0.5.5

2025년 12월 22일

0.5.52025년 12월 22일

Arch Linux Support & Windows Fixes

Arch Linux support, improved link handling on Windows, and a What's New button in Settings.

Highlights

Arch Linux support, thanks to Hong Lan
Fixed links not opening on Windows
What's New button in Settings

Added

Arch Linux package support
What's New button in Settings → Info section

Fixed

Links not opening on Windows
Remote LLM configuration with custom provider only

Platform

Linux: Added AppImage standalone executable

0.5.4

2025년 12월 20일

0.5.42025년 12월 20일

Linux Packaging & Graph Reliability

Linux packaging improvements with AppImage and RPM support, plus knowledge graph reliability enhancements.

Highlights

AppImage support for most Linux distributions
RPM packages for Fedora, RHEL, CentOS, openSUSE
Orphaned entity cleanup in knowledge graph
Bi-temporal relationship tracking

Added

RPM packages for Red Hat-based distributions
Orphaned entity cleanup tool
Bi-temporal fields for relationships

Improved

Community detection algorithm reliability
Entity alias matching and deduplication
Faster installation

Fixed

LLM provider handling for Gemini and Anthropic
AI Now tool call display
Linux AppImage archive paths

0.5.3

2025년 12월 18일

0.5.32025년 12월 18일

nmem CLI & New LLM Providers

New command-line interface for AI agents and power users, GitHub Copilot and DeepSeek R1 support, plus expanded conversation import options.

Highlights

nmem CLI for terminal-based memory access
GitHub Copilot integration (no API key needed)
DeepSeek R1 with visible reasoning traces
ChatGPT & ChatWise ALL historical conversations import

Added

nmem CLI with memory search, thread management, markdown import
GitHub Copilot as LLM provider (uses existing subscription)
DeepSeek R1 reasoning model with extended thinking
Custom OpenAI-compatible endpoint support
ChatGPT JSON export import
ChatWise conversation import
Bulk file import, import ALL historical conversations from ChatGPT & ChatWise(more upcoming)
Message selection when distilling threads
Source thread lookup for any memory
Global keyboard shortcut customization

Improved

Obsidian tag extraction with Unicode support
Wikilink parsing (header links, escaped pipes)
Thread full-text search
Automatic recovery from errors

Platform

macOS: Fixed parallel inference on Apple Silicon
Windows: Fixed orphan process cleanup
Linux: Fixed blank screen on Arch Linux

0.5.2

2025년 12월 9일

0.5.22025년 12월 9일

Deep Research Hotfix

Fixed

Deep Research output parsing
Research renderer confidence level handling
Linux update download flow

0.5.1

2025년 12월 7일

0.5.12025년 12월 7일

Stability Hotfix

Fixed

Parallel inference on Apple Silicon
AI Now on Windows Store version
Linux script path detection

0.5.0

2025년 12월 5일

0.5.02025년 12월 5일

AI Now & Bi-Temporal Memory

AI Now personal assistant, bi-temporal memory with cognitive-science decay model, Linux desktop support.

Highlights

AI Now: personal AI that knows you and your files
Bi-temporal memory with intelligent decay
Linux desktop support (Beta)
Up to 10x faster knowledge graph extraction

Added

AI Now: Personal assistant with memory, files, and app integrations
Notion, Obsidian, Apple Notes integrations for AI Now
Deep Research with multi-agent web + memory search
Document analysis (Excel, CSV, Word)
Presentation generation with PowerPoint export
Bi-temporal memory (event time vs record time)
Memory decay model inspired by cognitive science
Temporal intent detection for natural queries
Auto-update system with progress tracking

Improved

Knowledge graph extraction up to 10x faster
Fast search under 100ms
Deep search up to 5x faster
Graph visualization smoothness

Platform

Linux: DEB packages for Ubuntu/Debian
Linux: RPM packages for Fedora/RHEL
Linux: AUR support for Arch Linux
Linux: Dedicated launcher UI

0.4.11

2025년 11월 18일

0.4.112025년 11월 18일

Light Mode

Complete light mode support with theme-aware visualizations.

Highlights

Light mode across all views
Entity and community search
Windows startup improvements

Added

Light mode with theme toggle in Settings
Entity search in fast search
Community search across graph
Label filtering in fast search

Improved

Theme-aware graph visualization
LLM availability error reporting
Settings UI with tooltips

Fixed

Windows launcher startup flash
Windows console window blink

0.4.10

2025년 11월 15일

0.4.102025년 11월 15일

Windows Beta

Windows support arrives with all core features working.

Highlights

Windows Beta release
Light mode theming
Optimized Deep Search (~6 seconds)

Added

Windows Beta with full feature support
Light mode with system theme integration
Cross-platform embedding support

Improved

Deep Search optimized to ~6 seconds
Intent detection accuracy
Process cleanup on Windows

0.4.6

2025년 10월 29일

0.4.62025년 10월 29일

DeepChat & MCP Improvements

Native DeepChat integration and improved MCP tool definitions.

Highlights

Added

Improved

MCP tool argument prompts

0.4.2

2025년 10월 24일

0.4.22025년 10월 24일

Thread Search & Claude Code

Search across all threads and import from Claude Code and Codex.

Highlights

Added

Thread search across all conversations
Claude Code plugin for conversation import
Codex integration with thread import

Improved

Thread view performance
Code block rendering for large threads

0.3.1

2025년 10월 12일

0.3.12025년 10월 12일

Fast Search & Remote LLM

Quick memory lookup and cloud LLM provider support.

Highlights

Fast search mode for quick lookup
Remote LLM provider support

Added

Fast search mode
Remote LLM provider support

Improved

LLM-based intent detection

0.2.4

2025년 9월 22일

0.2.42025년 9월 22일

License & Updates

Lifetime License activation and update checker.

Added

0.2.0

2025년 8월 7일

0.2.02025년 8월 7일

Alpha Release

First public alpha with core memory management, knowledge graph, and MCP integration.

Highlights

Added

Threads view for conversation management
Memories view with knowledge cards
Knowledge graph visualization
MCP server for Claude, Cursor, and other AI tools
Browser extension (Nowledge Mem Exchange)
Search launcher (⌘⇧Space)
Semantic search with embeddings
Graph multi-selection and lasso select
Native macOS app