Skip to main content

Changelog

The meaningful changes in Nowledge Mem, written for people using it every day.

Latest release
0.10.121
19. Nov. 2026
0.10.122NextComing next

Quicker Graph previews and more ways to work with Artifacts

Start exploring the Graph sooner, download Artifact folders on supported servers, and keep conversations together when moving them across Spaces.

Highlights
  • Download an entire Artifacts folder as a ZIP that preserves its structure. AI Now and the browser extension can save files directly into a folder, and folder changes from other devices appear without a reload. These features are available when your connected server has Artifact folders enabled.
  • Explore an interactive Graph preview while the full overview loads. The completed Graph preserves your selection and view.
  • Read the first page of Artifact text while related details load. The Artifacts list also fetches the next page before you reach the end, and keeps your filters when a page needs a retry.
Added
  • Download an entire Artifacts folder as a ZIP that preserves its structure. AI Now and the browser extension can save files directly into a folder, and folder changes from other devices appear without a reload. These features are available when your connected server has Artifact folders enabled.
  • Use Sync settings to exclude temporary or automated workspaces from local session discovery and automatic capture. You can still import those sessions manually.
Improved
  • Explore an interactive Graph preview while the full overview loads. The completed Graph preserves your selection and view.
  • Read the first page of Artifact text while related details load. The Artifacts list also fetches the next page before you reach the end, and keeps your filters when a page needs a retry.
  • On narrow AI Now windows, the model picker gives model names more room. On Windows and Linux, Skill previews leave the window controls accessible when you resize the pane.
Fixed
  • Moving multiple conversations between Spaces now keeps each conversation and its messages together through interrupted moves and restart recovery.
  • Choose Keep existing history when reviewing a Codex archive. Mem remembers the decision for that archive revision across watcher restarts.
  • Pressing Enter to confirm Chinese, Japanese or Korean text composition no longer submits the affected forms. Captured tables also preserve line breaks inside cells in conversations and AI Now.
  • Processing shows the last recorded reply check for supported distillation failures. Knowledge Graph errors provide more specific recovery details when the cause is known.
Platform
  • The experimental HawDB backend now uses HawDB 0.35.1. Preparing that backend avoids holding Search open during its comparison checks.
0.10.121Latest release19. Nov. 2026

Organize Artifacts and keep distillation moving

File documents into folders, find them from connected agents, and improve conversation imports, recovery and network compatibility.

Highlights
  • Organize Artifacts into folders, move documents by dragging or using Move to folder, and upload straight into a folder. Deleting a folder keeps its documents by default. Folder controls appear when the connected Mem server supports them.
  • Connected agents can find documents with artifact_search. Artifact tools and the nmem artifacts command use the same name as the app; existing library and sources commands and legacy tool calls keep working. Previously approved tools may ask for approval once under their new names.
  • Eligible failed automatic distillations can resume once with an adjusted request. Recovery better recognizes Ollama and Gemini settings, and slow post-call checks no longer discard paid distillation work merely for exceeding a foreground latency budget.
Added
  • Rebuild a blocked AI Now conversation's context in the same conversation. Saved history and drafts stay available, and rebuilding does not send a message automatically.
  • Local and self-hosted operators can enable official Slack JSON ZIP imports with NMEM_SLACK_EXPORT_IMPORT_ENABLED=1. Imports preserve threads and available attachment links; hosted Cloud direct ZIP imports are unchanged.
  • Organize Artifacts into folders, move documents by dragging or using Move to folder, and upload straight into a folder. Deleting a folder keeps its documents by default. Folder controls appear when the connected Mem server supports them.
  • Connected agents can find documents with artifact_search. Artifact tools and the nmem artifacts command use the same name as the app; existing library and sources commands and legacy tool calls keep working. Previously approved tools may ask for approval once under their new names.
Improved
  • Home notifications show their creation date and time in your app language and local timezone. The combined Space view is now simply named All.
  • When experimental HawDB is the Graph authority, Space observations and Thread and Source counts use that selected storage.
  • Label consolidation uses stricter duplicate guidance to distinguish broader and narrower terms, negation and status, results and activities, and roles and domains.
  • Experimental HawDB is updated to 0.34.0. Libraries written by a newer app version receive a specific message, and search-index errors no longer take the whole HawDB runtime out of service.
  • Experimental HawDB preparation avoids repeated full search-index rewrites for bookkeeping changes. Its final receipt check no longer reserves memory for the entire search index.
  • Distilling long conversations avoids repeatedly preparing earlier message packets that have already been accepted.
  • When a later conversation changes only part of a multi-fact Memory, its unchanged details remain searchable and linked to the update. Full replacements still replace the earlier Memory.
  • App imports can distill several conversations concurrently with a cloud model. Local and self-hosted models process one at a time; Task Center shows each running task with its own Stop button.
  • Chinese, Japanese and Korean text uses regional system fonts, while Latin text keeps the interface font. Chinese and Japanese emphasis uses weight, and punctuation follows regional forms; middle dots in names remain half width.
  • The Memories toolbar fits on one row, and list footers show totals. Memories supports 20, 50 or 100 items per page. Desktop Search and Settings show their keyboard shortcuts.
  • Auto-Sync avoids repeatedly parsing earlier turns in growing Codex conversations. Session imports also avoid duplicate image inputs when matching raw and Event records use different message IDs.
  • Conversation rows gather sharing, pinning and deletion in one menu and show a single browse total. Search keeps its Space selector, query and Clear button together, and Home marks the end of a fully loaded timeline.
  • With experimental HawDB selected, Label and AI identity usage reads follow that storage. Thread message mirroring also preserves recovery order, including empty conversations and writes retained after a mirror failure.
Platform
  • Desktop update checks and CLI Cloud Bridge connections work on TCP-only networks and reuse connections. Interrupted Bridge writes retain their original operation for recovery.
Fixed
  • Native Bridge can wait briefly for a busy session to become available, without resending your original operation.
  • Local search failures retain their specific reason and show clearer recovery guidance, including when a search-index rebuild needs support rather than an automatic retry.
  • Conversation distillation and source-derived Memory writes no longer stall while saving results. Saved progress remains available for retry when a conversation grows, and uncertain duplicate-review outcomes retain guidance against paying again.
  • Experimental HawDB clearly identifies read-only storage and transactions that exceed size or operation limits, with guidance to contact support before retrying.
  • Missing or inaccessible Memory, Artifact, Crystal and Entity details show a persistent in-app message with a working return action. Connection failures remain distinct.
  • Pressing Escape in a reopened confirmation keeps Team settings open and returns focus to the control that opened the confirmation.
  • Memory updates return the saved content, full ID and a revision for optional conflict protection. Outdated conditional edits are rejected without overwriting newer changes.
  • Memory and conversation references open their target or show a persistent unavailable message with a way back. Historical Thread dates and recent relative times stay visible while reading.
  • App controls and surrounding UI no longer select text during a drag. Memory and document content, fields and marked copyable values remain selectable. The macOS sidebar toggle stays aligned with the window controls.
  • OAuth connections handle client-registration expiry and token revocation more reliably. Disconnecting a local MCP plugin prevents delayed authorization or refresh responses from reconnecting it.
  • Memory lifecycle filters keep counts and pagination aligned, and imported Memories preserve supplied importance and confidence metadata when top-level scores are omitted.
  • Eligible failed automatic distillations can resume once with an adjusted request. Recovery better recognizes Ollama and Gemini settings, and slow post-call checks no longer discard paid distillation work merely for exceeding a foreground latency budget.
  • Memory review rechecks possible state changes before reporting a conflict. New conversation Memories use the times of their cited messages, and distillation can correct an invalid duplicate target once within its task budget.
  • Search diagnostics distinguish configured embedding settings from the embedder running in the backend, including changes that require a backend restart. Checking status does not load a model or contact its provider.
  • Plugin installs and updates use registered marketplace commands and validate the full update command. Native Skill reads stay within their result and time limits, with clearer failure reasons.
  • Distillation can recover from a reasoning-only reply when the adjusted request fits the available time. Later runs on a route that exhausted its first reasoning budget can start with more room.
  • Distillation accepts an OpenAI-compatible endpoint already saved without an API key, matching chat behavior. If a saved key cannot be decrypted on this device, the app asks you to enter it again.
  • Codex auto-sync stops repeatedly importing an unchanged conversation snapshot after a source-history conflict. The conflict stays visible, and syncing retries when the file changes or during its daily recheck.
  • Editing Memory metadata preserves the original source timestamps used to order related memories. Full Data Transfer restores retain archived timestamps.
0.10.11915. Nov. 2026

Clearer Memory history and more reliable connections

Keep your Memory corrections current, find the next step in Ontology, and improve graph search, conversation imports and Bridge connections.

Highlights
  • Memory review and cleanup keep your explicit correction of an assistant's statement current, even if the assistant later repeats the rejected statement.
  • Use nmem m evolves repair-direction to preview, apply or undo corrections to older version links saved in the wrong direction. Each repair appears on the Timeline.
  • The Graph Ontology panel shows the next action for your current setup and whether your types are being used on new content. Settings shows the same status, including when the background model cannot apply them.
  • Bridge setup can find Agent Contexts retained by connections connect, while preserving the existing Agent identity and registration recovery.
Added
  • Memory search shows when a newer Memory updated an older one, helping connected AI tools and the CLI identify the current statement.
  • Use nmem m evolves repair-direction to preview, apply or undo corrections to older version links saved in the wrong direction. Each repair appears on the Timeline.
Improved
  • Choose reasoning effort, including None, under Advanced for built-in OpenAI and Magpie connections when the model list does not declare the setting.
  • Session sync gives projects more scrolling space, clearer filtering and keyboard actions, and direct access to available source error details.
  • Label consolidation can skip a stale merge pair without failing the entire batch, with clearer results for completed and skipped changes.
  • App-hosted MCP browsing reads Sources and Libraries in bounded batches instead of loading the entire library at once.
  • When experimental HawDB is the Search authority, fast Memory search includes exact vector candidates and correctly handles default Spaces and content-type filters.
  • Automatic relationship review considers highly ranked vector matches across languages while keeping a bounded candidate budget.
  • The Graph Ontology panel shows the next action for your current setup and whether your types are being used on new content. Settings shows the same status, including when the background model cannot apply them.
  • Clear Graph search with the field's clear button, Escape in the field, or by deleting the query to return to the overview.
  • Label cleanup cards remain visible across Space filters and include a shortcut to manage labels.
Fixed
  • Bridge setup can find Agent Contexts retained by connections connect, while preserving the existing Agent identity and registration recovery.
  • Codex conversations with images keep a stable parsed history as new turns arrive, preventing a repeated caption from splitting an earlier image and blocking later sync.
  • Queued Memory edits respect revoked access, and overlapping background classification preserves confirmed status and metadata changes.
  • Conversation saves and Space moves preserve confirmed work when a later step fails, recover interrupted moves without overwriting newer changes, and keep partial-import message counts accurate.
  • Direct conversation saves in a personal workspace use the AI Identity default Space when no Space is explicitly selected.
  • Automatic relationship review no longer mistakes an edited Memory original creation time for the date of its current statement.
  • Skill bundle imports reject filenames that become identical after Unicode normalization, avoiding conflicting files on supported filesystems.
  • Background tasks through Magpie can retry an answerless thinking reply that exhausts its output allowance. Structured tasks through Magpie and OpenRouter also receive JSON mode when supported.
  • DeepSeek V4 and MiMo connections consistently receive the selected thinking-off setting across supported request formats.
  • Failed background-run reports include the recorded diagnostic codes to help support investigate the original run.
  • Team Lite preserves existing members, invitations and keys when damaged settings prevent changing its enabled state.
  • App distillation can ask the model to correct the specific missing or misplaced field in a structured reply, improving recovery from invalid reply shapes.
  • Memory review and cleanup keep your explicit correction of an assistant's statement current, even if the assistant later repeats the rejected statement.
  • Bulk import progress stays attached to the current attempt, and partially completed imports keep their errors visible.
  • Bridge sessions now work through authenticated local servers and Access Anywhere connections.
  • Fixed full search-index rebuilds stopping on large Libraries when Source Chunk reads exceeded their budget.
  • MCP Memory-neighbor and Skill reads report resource or time-limit errors clearly instead of generic failures or invalid-input errors.
  • MCP conversation sync rechecks the destination Space after waiting to save, preventing a conversation from being saved into a Space deleted during that wait.
  • The Bub connector introduction correctly identifies Bub as a Python agent harness.
  • Desktop update checks include HTTP/3 connection and QUIC compatibility fixes.
Platform
  • Experimental HawDB storage uses engine 0.31.0. Rejected writes, such as duplicate keys, return an operation error without taking the HawDB runtime out of service.
  • Experimental HawDB preparation handles libraries without older session schemas, preserves Legacy column defaults, and checkpoints long imports while reclaiming completed preparation captures.
  • Experimental HawDB gives distinct guidance for constraint violations, schema mismatches and operations whose publication result is uncertain.
0.10.11812. Nov. 2026

More reliable background AI and local storage

Keep background AI choices and settings intact, simplify AI Now paths, and improve repair reliability in HawDB Labs storage modes.

Highlights
  • In local Labs Identity graph, summarize a conversation into editable identity cards. Choose which drafts to save and link agents or roles to an AI Identity; nothing is saved until you confirm.
  • AI Now uses application storage for new installations while keeping existing files in place. Windows file paths are easier to read.
  • Background AI keeps the selected model and funding account through retries and PDF OCR, and applies spending limits to the route each task uses. Subscription and local routes retain paid usage history without being blocked by limits that do not apply to them.
  • Conversations keep their precise activity time in the list. Records with only a calendar date keep that date instead of appearing many hours old.
Added
  • In local Labs Identity graph, summarize a conversation into editable identity cards. Choose which drafts to save and link agents or roles to an AI Identity; nothing is saved until you confirm.
Improved
  • Desktop and self-hosted MCP connections can save, search and read Memories in the selected Space. Conversation search and reading are also available while legacy Graph storage is authoritative.
  • Locally extracted Memories follow the conversation's language when no output language is set in your Profile. An explicit Profile language still takes priority, and names keep their original spelling.
  • Local relationship and cleanup reviews retain usable suggestions when a model returns an invalid one. Partially reviewed Memories receive bounded follow-up reviews, and dismissed contradictions stay dismissed.
  • Local Memory relationship reviews use shared entities, labels and conversations. Older conversations imported later become predecessors of newer statements; statements with no clear order remain review cards.
  • Choose low, medium or high reasoning effort for native Gemini 3.8 Flash in provider settings. Structured background tasks can use its lowest supported effort.
  • Locally extracted Memories resolve phrases such as "yesterday" to dates using the original message time and your timezone, while retaining dates mentioned in passing.
  • Remote MCP connections support registered client secrets when enabled by the operator. Existing public-client connections keep their current setup.
  • Local embedding repairs make room for waiting foreground requests between batches.
  • Experimental HawDB storage uses engine 0.30.0. Older experimental datasets may need separately verified preparation in a new location; existing data is preserved and incompatible startup is refused.
Fixed
  • AI Now uses application storage for new installations while keeping existing files in place. Windows file paths are easier to read.
  • Background AI keeps the selected model and funding account through retries and PDF OCR, and applies spending limits to the route each task uses. Subscription and local routes retain paid usage history without being blocked by limits that do not apply to them.
  • Settings keep limits returned by older servers and preserve unsaved edits when a conflicting or outdated response arrives, until you explicitly reload.
  • In HawDB Labs storage modes, conversation repair can consolidate obsolete conversations whose IDs differ only in letter case, while retaining retry recovery.
  • In HawDB Labs storage modes, Identity activity and profiles use the selected storage consistently. Saved changes and pending cleanup work are retained when a later verification fails; unavailable evidence is reported rather than treated as empty.
  • In HawDB Labs storage modes, Skill publication keeps the saved procedure and recovery files when the final confirmation fails. Memory vector search also filters Memory records within the same read snapshot.
  • OpenCode automatic sync avoids repeated imports of legacy JSON already covered by its active database. If sync stops partway through a database, it keeps completed conversations and resumes unfinished work after restart.
  • Manual and automatic local distillation share progress for each conversation revision, avoiding repeated extraction of the same content while allowing new messages to be processed.
  • Onboarding keeps failed imports, recovery guidance and Retry import available on both history-import screens. Retrying preserves successful imports, and finished imports no longer appear to await an AI read before AI is configured.
  • Conversations keep their precise activity time in the list. Records with only a calendar date keep that date instead of appearing many hours old.
  • Skill creation keeps saved Skills visible when background setup cannot be confirmed. Retrying keeps review-only Skills inactive, enrichment retains saved examples, and completed updates stop showing Updating.
  • Successful Source refreshes, reparsing, indexing and status changes stay successful when storage compaction is unavailable.
  • Memory type reviews in the CLI distinguish queued, already queued, skipped and unconfirmed requests, with recovery guidance when admission cannot be confirmed.
  • Dragging the Skill helper title moves its desktop window.
0.10.11710. Nov. 2026

Clearer Controls and More Useful Recall

Simpler settings, readable dated recall, and background processing with fewer interruptions.

Highlights
  • A new first-run guide helps you choose AI, connect tools, select sessions to import and get a first answer from your history. Imported sessions are searchable while Memories are extracted in the background.
  • Long conversations can retain more useful Memories as their content grows, while short exchanges stay concise. Focused extraction keeps its one-Memory limit.
  • Conversation search quotes the matching passage from the original message, with its timestamp when available. Chinese excerpts are readable, and CLI JSON retains passages and conversation dates so agents can trace the evidence.
Added
  • A new first-run guide helps you choose AI, connect tools, select sessions to import and get a first answer from your history. Imported sessions are searchable while Memories are extracted in the background.
Improved
  • Skills and Tools have separate tabs, with easier-to-read Skill descriptions and clearer activation and verification status.
  • Skill details show the current version beside the name; proposed updates stay separate until you apply them.
  • Long conversations can retain more useful Memories as their content grows, while short exchanges stay concise. Focused extraction keeps its one-Memory limit.
  • Conversation search quotes the matching passage from the original message, with its timestamp when available. Chinese excerpts are readable, and CLI JSON retains passages and conversation dates so agents can trace the evidence.
  • Settings puts Rules in one editor with a clear scope. AI Models starts with search and memory processing choices, while Providers focuses on sign-in and connections.
  • Ask about the whole graph without selecting anything first. Memory history opens over the Tree so you can keep your place.
  • Home shows conflicting memories side by side with their dates and groups similar pairs into one compact row. Background issues use clearer wording and a suggested next action.
  • Subscription or coding-plan sign-ins and local models no longer pause background work under a dollar limit. For your own API keys, the default safety limits are USD 15 a day and USD 40 a month; custom limits are preserved.
Fixed
  • GitHub Copilot models that require the Responses API can answer and use tools. Connection checks verify the selected model and explain when it cannot be used.
  • An unreadable citation or malformed point in an AI reply no longer discards the rest of a conversation’s Memory extraction. Replies with format errors can be corrected within the existing work budget.
  • Bub plugin setup and updates use the Python environment that actually holds the plugin. For development or system-managed environments, the Integrations card explains the manual command instead of trying unrelated Pythons.
  • General and Memory spaces settings use the available pane width, keeping controls and descriptions readable beside an expanded sidebar.
  • License status distinguishes an unavailable stored-Memory count from a count of zero.
  • Slow startup storage checks warn instead of rejecting valid Content data solely because inspection takes more than one second.
  • Manual conversation distillation uses the output language selected in your Profile. If the Profile cannot be read, you can retry without losing already accepted work.
  • Bulk move and delete preserve the selected source Space and reject selections above the 5,000-record limit before changing data. Incomplete moves keep your selection available for review and retry.
  • Compact Tree and Library graphs keep relationship lines visible and the type legend readable. Parallel relationships retain their labels, and metadata no longer replaces canonical graph identities.
  • Plus and Team show their unlimited device allowance correctly. Sign-in can recover from an older device-limit session, and account refreshes no longer interrupt completed Personal or Team sign-in.
  • Background work waits and resumes automatically when an AI provider rate-limits requests. An empty balance still shows a top-up action.
0.10.1167. Nov. 2026

Background Progress and Smoother Browsing

Track session sync and Memory extraction, resize the sidebar, and browse with steadier startup and clearer lists.

Highlights
  • The Local App keeps Memory extraction progress visible in Home, the sidebar and Threads, with queued, running and waiting states and links to the conversations.
  • Memories, Threads, Artifacts, Skills and Identities keep their title and toolbar visible while only the list scrolls, with more room in wide windows.
  • Large pending Thread deletion backlogs no longer prevent the App from opening.
Improved
  • Improved Knowledge Graph extraction for long Memories and large batches.
  • Memories, Threads, Artifacts, Skills and Identities keep their title and toolbar visible while only the list scrolls, with more room in wide windows.
  • Home loads the graph renderer when you open its reader or graph panel, reducing the code loaded up front.
  • Slack Threads with multiple speakers show each sender’s name and avatar, making conversations easier to follow at any window size.
  • Update checks can use HTTP/3 while retaining the existing HTTP client as a fallback for network failures.
  • Drag the sidebar edge to resize it smoothly; narrow windows keep the icon rail available.
  • With the Identity graph Lab enabled, changes to the people or Agents grouped into an AI Identity offer a new summary; review each field and keep or replace it.
  • Desktop has a DNS over HTTPS switch in Settings > General > Network for its hosted Cloud connection. It starts off and can be turned off again to restore system DNS.
  • Memory health review clearly distinguishes changing unconfirmed notes to background notes from leaving them unchanged, and reports how many notes are not shown.
  • The Local App keeps Memory extraction progress visible in Home, the sidebar and Threads, with queued, running and waiting states and links to the conversations.
  • When automatic session sync is enabled, Threads shows the last check and sources needing attention, with links to sync settings.
  • AI Now no longer offers the Didi plugin. Existing tasks keep their other plugins; Didi-only tasks stay readable and offer a new conversation to continue.
Fixed
  • Large pending Thread deletion backlogs no longer prevent the App from opening.
  • Quitting or restarting Desktop interrupts local model downloads safely, so unfinished downloads can be retried after reconnecting.
  • Local installations that have only used Default Space repair eligible legacy Entity and relationship ownership in the background; ambiguous data stays unchanged.
  • AI Now uses the task’s Space when opening or searching files, counting Memory references and building Wiki pages.
  • Clicking outside an open Memory closes its reader; editing keeps it open, and choosing another Memory switches directly to it.
  • On Windows and Linux, the AI Now header stays clear of the window controls.
  • Expanded error details use the full notice width, with action buttons wrapping in narrow dialogs.
  • The Remote workspace menu distinguishes your server’s name from its connection details.
  • Model pickers expose their popup relationship to assistive technologies.
  • Memory reviews explain when a suggested relationship would form a loop and keep the review visibly pending.
  • Slow Library scans have time limits, and failed upload-version checks no longer leave unused files behind.
  • CLI commands keep queued, skipped and refused work distinct when reporting review, detection and Skill compilation results.
  • AI Now explains when a Memory is outside the task’s Space instead of treating its ID as incorrect.
  • Upgrades normalize legacy Memory lifecycle states before serving lists and counts, so archived or removed items are classified consistently.
  • On macOS, the collapsed sidebar button follows the actual window-button position to avoid overlap. The icon rail now shares the page background.
  • Home can find conversations imported through the API for Memory extraction, including mixed sources, within the import’s Space.
  • Copying connection prompts and hosted MCP addresses is more reliable in Desktop; if copying fails, the generated text stays available for manual copying.
  • Moving a conversation to another Space no longer reports failure after the move has already been saved.
0.10.1156. Nov. 2026

Identity Tools and Easier Everyday Work

Download Artifacts together, manage Spaces and Connectors in Settings, and work with clearer conversations, reviews and optional identity tools.

Highlights
  • Download several selected Artifacts as one ZIP, with original files or Markdown for text-only items. Any files that cannot be included are reported.
  • Manage and create Spaces from a dedicated Settings page, also reachable from the account menu.
  • With the optional Identity graph Lab enabled on Desktop, combine discovered profiles into an AI Identity, review a source-based identity draft, and ask suggested questions from graph details. AI Now uses the linked identity card from the conversation’s Space.
Added
  • Download several selected Artifacts as one ZIP, with original files or Markdown for text-only items. Any files that cannot be included are reported.
  • With the optional Identity graph Lab enabled on Desktop, combine discovered profiles into an AI Identity, review a source-based identity draft, and ask suggested questions from graph details. AI Now uses the linked identity card from the conversation’s Space.
  • Use nmem labs to inspect and change this Local App’s optional Labs settings. HawDB preparation and App restart remain explicit actions. Cloud and remote connections leave device settings unchanged.
Improved
  • Connectors now lives in Settings, after AI configuration. Tool logos, setup instructions and card actions are clearer at narrow window sizes.
  • Thread details can show thinking time and applied file changes when the imported conversation provides them. Reviews preserve existing titles and summaries instead of replacing them with prompt templates.
  • Home conflict reviews highlight differences and keep instruction drafts and replies visible. Recent processing runs are easier to distinguish from older recovery work.
  • Manage and create Spaces from a dedicated Settings page, also reachable from the account menu.
  • The sidebar identifies where your current Mem lives even when no account is signed in, and labels self-hosted servers by address.
Fixed
  • Local AI providers such as Magpie connect directly when a system or environment proxy is enabled. Remote providers keep their proxy settings.
  • Space merges remain successful after their data has committed, even if later cleanup needs attention. Local Space and MCP listings report Working Memory presence more accurately.
0.10.1145. Nov. 2026

Refreshed Reading Views and Simpler MCP Setup

Navigate your knowledge with clearer reading views, configure MCP in one place, and recover imports and loading states more reliably.

Highlights
  • Home, Memories and Threads have clearer navigation and reading panels. Browse Timeline history, keep Focus drafts across navigation, and open related records without losing your place.
  • Local MCP configuration and Remote MCP OAuth pairing now share the MCP page, with separate tabs and shared Live Traffic. Conversation imports live in Threads.
  • Deleting a conversation and importing it again now reviews and distills it again instead of treating it as already processed.
Improved
  • Home, Memories and Threads have clearer navigation and reading panels. Browse Timeline history, keep Focus drafts across navigation, and open related records without losing your place.
  • Tree, Graph and Wiki have dedicated navigation and contextual readers. Artifacts offers a searchable list, supported document editing and version history.
  • Local MCP configuration and Remote MCP OAuth pairing now share the MCP page, with separate tabs and shared Live Traffic. Conversation imports live in Threads.
  • Local session imports use less memory for status responses, and model replies with malformed markup require less background parsing work.
  • Connectors brings AI tools into one searchable list, with a detail view and a clear setup action for the tool you choose.
  • Context shows discovered people and agents with avatars, descriptions and aliases, with search, filters and paged loading. AI Now makes the AI Identity picker visible before your first Identity is created.
Fixed
  • Deleting a conversation and importing it again now reviews and distills it again instead of treating it as already processed.
  • Pending sign-in links no longer loop app navigation. Activity Calendar, Stats and Memories refresh recover from stalled reads, and Connectors navigation remains usable after LobeHub setup.
  • Provider setup explains when an API key is required before fetching models. Magpie on another computer asks for its gateway key before saving.
  • The Chinese all-types filter stays readable in narrow windows, and AI Now model and working-folder controls align with the message box.
  • Reloading the page keeps the Context tab you had open instead of returning to Preview.
  • Initial pending statistics show zero totals and empty activity, then update when observations arrive. Failed refreshes keep previously successful values.
  • Conversation storage repair distinguishes merged conversations from removed empty duplicates, keeps partial and pending results visible, and requests a fresh review when an earlier repair plan is stale.
  • Artifacts enables Markdown editing only when saving is supported. If that support changes during an edit, your draft remains available to copy and Cancel stays usable.
  • Managed-service status recovers after temporary failures and refreshes when you return to the app. Local Vault settings hide unrelated hosted-service status.
  • Chinese Focus briefings on Home use clearer spacing, and affected setup fields wait until Chinese text composition finishes before submitting.
Platform
  • The embedded database is updated to HawDB 0.27.0. Runtime admission remains off by default for datasets that have never enabled it.
0.10.1133. Nov. 2026

Clearer Skills and More Reliable Conversation Recovery

Continue Skill improvements in AI Now, inspect recorded AI usage, and import recovered conversations with fewer interruptions.

Highlights
  • Skills with AI Now editing available have a simpler detail view, a single improvement action and quick access to recent improvement conversations.
  • Inspect every AI usage source recorded by this Mem instance by time period and activity. Calls without reported token counts remain visible.
  • Previously recovered Codex archives, including compressed archives, can be imported again without duplicate messages when their complete source history is verified.
Improved
  • Skills with AI Now editing available have a simpler detail view, a single improvement action and quick access to recent improvement conversations.
  • Inspect every AI usage source recorded by this Mem instance by time period and activity. Calls without reported token counts remain visible.
  • The built-in local model uses a working window sized to your computer's memory for conversation distillation. CPU-only setups explain the expected slowness and available model alternatives.
  • Conversation and file imports avoid additional temporary message copies while preserving imported content, order and retry behavior.
  • Access Anywhere clearly warns that Quick link addresses change after every restart and points to persistent connection options.
Fixed
  • AI Now refreshes a saved Skill proposal so Apply and Dismiss appear immediately. The sidebar editor uses the available height and opens full details while retaining your draft.
  • Recovering AI Now history carries the visible conversation and unsent request into an editable draft without automatically resending it.
  • Previously recovered Codex archives, including compressed archives, can be imported again without duplicate messages when their complete source history is verified.
  • Reading messages with unusually long Markdown runs stays more responsive while preserving formatting and complete copy and export content.
  • Conversation distillation skips legacy developer instructions while preserving the original saved transcript.
  • Distillation can recover replies that run out of output space on slower providers with a fresh request window; recovery sizing uses reported output evidence.
Platform
  • The embedded database is updated to HawDB 0.25.0. Runtime admission remains off by default for datasets that have never enabled it.
  • The backend's live API documentation reports the version of the running backend.
0.10.1121. Nov. 2026

Safer Conversation Actions and Lighter Imports

Keep conversation actions tied to the selected saved history, import large archives with less memory, and preserve model settings through edits.

Highlights
  • Conversation reads, exports and handoffs retain the selected saved conversation and Space when public IDs overlap. Ambiguous references offer recovery instead of choosing another row.
  • Proma and large local Agent Session imports use less peak memory while preserving complete messages, selection order and failed-session retry.
  • Access Anywhere no longer mistakenly blocks browser-based MCP clients' public OAuth discovery, registration and token requests. Protected routes retain their origin checks.
Improved
  • Proma and large local Agent Session imports use less peak memory while preserving complete messages, selection order and failed-session retry.
  • Large ZIP and direct-file imports stop at their read limits and report oversized content. Healthy parsed content remains available where supported, and processed payloads are released sooner.
  • Claude account imports can continue with healthy conversations and project documents when profile memories are oversized. Import previews distinguish available content from completed imports.
  • Distillation failure details identify the rule and draft or judgment that failed validation, without recording the model reply text.
  • With Identity Graph enabled, removing a profile hides it from active lists while keeping saved Memories and confirmed references intact.
  • Antigravity subagent conversations sync as separate Threads when their recorded delegation can be read. Delegated tasks remain system messages, and unavailable lineage is explained without an unhelpful retry prompt.
  • Access Anywhere provides a connector file with the sign-in and token addresses that hosted agents need, reducing setup errors caused by manually retyping long URLs.
Fixed
  • Access Anywhere no longer mistakenly blocks browser-based MCP clients' public OAuth discovery, registration and token requests. Protected routes retain their origin checks.
  • Conversation reads, exports and handoffs retain the selected saved conversation and Space when public IDs overlap. Ambiguous references offer recovery instead of choosing another row.
  • Conversation actions recheck the saved identity before changes. Retrying a partial bulk deletion keeps excluded conversations out of the retry.
  • Moving a conversation retires pending actions for its previous Space. Reopening active distillation retains its confirmation, task and result tracking.
  • Stopped search repairs stay tied to the correct saved conversation, with read-only message inspection in its original Space.
  • Assigning Magpie models to Fast decisions preserves saved provider settings and unsaved edits. Reasoning choices and an independent Default model survive unrelated saves; conflicting writes share pending controls.
  • Stats and onboarding retain their last successful values when background refresh fails, and recover without a disruptive retry prompt.
  • Desktop and self-hosted local Spaces show Delete and Merge actions again. Contextual Edit and Delete labels are shorter and clearer.
  • Threads keeps keyboard focus on the chosen control after navigation.
  • Process diagnostics show CPU measurements as unavailable when a translated Mac process has unverified clock units, while retaining process identity and memory readings.
  • macOS updates started from this version tolerate unrelated SMB mount warnings during the backend port check, while real process-inspection failures still stop the update.
  • Full-text index rebuilds check available disk space before replacing the final index. If space is insufficient, the existing index is kept and the rebuild can be retried after freeing space.
  • Switching AI modes and restoring settings preserve working model connections and saved selections. Incomplete routes remain visible with repair guidance and skipped-connection warnings.
  • Connection-status refreshes no longer overwrite the saved OAuth authorization state.
  • Memory Health shows the recorded Keep for now result when the server confirms that Memories were kept, even if its optional cooldown flag is absent.
  • Timeline evidence reviews close when a refresh confirms that all selected decisions were already recorded, without submitting those decisions again.
  • Local Working Memory and AI startup context served by the app no longer treat old memory-and-thread count summaries as the current briefing. Genuine briefings and additional content remain readable, including in past Timeline entries.
  • Codex shell commands identified by the host no longer become conversation content or Thread titles. Ordinary messages quoting shell-command output remain readable.
  • Skill evidence updates preserve timestamps when counting the days on which evidence occurred.
  • Invalid AI provider settings requests return consistent JSON error details.

Ready to try Mem?

Download the app, connect your AI tools, and keep useful context where you control it.