Install the community DeepSeek Harness bundle with Context Bundle, prompt-time recall, Mem MCP tools, and turn-end thread capture.

For your agent

Paste this into DeepSeek Harness or another agent helping with setup:

Read https://mem.nowledge.co/SKILL.md and follow the instructions to install or update Nowledge Mem for DeepSeek Harness. Use the community DSH bundle, verify with nmem status and the Context Bundle or Working Memory check, then run one short DSH session and confirm nmem t list --source deepseek-harness shows the captured thread.

DeepSeek Harness (dsh) is a plugin-first agent harness. Nowledge Mem uses the host's native Cordis lifecycle: startup context and recall are injected during agent/pre-step, Mem tools are bridged through DSH's MCP client, and real DSH transcript events are imported after turns finish.

This is a community plugin package published from nowledge-co/nowledge-mem-deepseek-harness. The repo is tagged dsh-plugin so DeepSeek Harness users can find it through the ecosystem convention.

Before You Start

Nowledge Mem is running locally, or you have a reachable Nowledge Cloud / remote Mem endpoint
DeepSeek Harness is installed and dsh works
nmem is on PATH in the same environment that launches DSH

nmem status
dsh --help

Setup

Install the community bundle into the DSH web profile:

dsh plugin --profile web add github:nowledge-co/nowledge-mem-deepseek-harness

Restart DeepSeek Harness:

dsh web

If you are installing from a local nowledge-co/community checkout instead of the published package:

dsh plugin --profile web add ./nowledge-mem-deepseek-harness-plugin

Remote Mem

For Nowledge Cloud, Access Anywhere, self-hosted, or authenticated localhost, configure this machine once:

nmem config client set url https://your-server
nmem config client set api-key nmem_your_key
nmem config mcp show --host deepseek-harness

For the DSH bundle itself, the most direct environment is:

export NMEM_MCP_URL="https://your-server/mcp"
export NMEM_API_KEY="nmem_your_key"

Then restart dsh web.

Verify

Ask DSH:

Search my Nowledge memories for recent decisions about this project.

Success means DSH can see the Mem MCP tools or the native recall message and returns results or a clean empty result.

Then run one short DSH session and check:

nmem t list --source deepseek-harness

You should see a thread whose id begins with deepseek-harness-.

What The Bundle Uses

DSH surface	Nowledge Mem behavior
`agent/pre-step`	Injects the Context Bundle once per DSH session and adds prompt-time recall when the prompt looks like continuation, prior-decision, release, regression, connector, or plugin work.
`@deepseek-ai/dsh-mcp-client`	Exposes Mem tools under the `mcp__nowledge_mem__...` namespace.
`session/event`	With plugin 0.1.7, saves human input and assistant reply text after `turn/end` through `nmem t import --source deepseek-harness`. Tool-call steps, tool results, reasoning and injected context are excluded.

After updating to 0.1.7, restart DSH and try a new short conversation that uses a tool. Open its Thread in Mem: your question and the assistant's answer should appear without the tool trace or runtime context. The same capture rules apply to a Mem endpoint reached through the App, Access Anywhere or Nowledge Cloud. Updating does not remove messages already imported by an older plugin. If an older session reports an import conflict, contact support before deleting or re-importing its Thread.

Known Limits

DeepSeek Harness is in developer preview and may ship breaking plugin API changes.
The connector captures new sessions while the bundle is active; historical DSH backfill is not claimed yet.
DSH's MCP bridge currently exposes tools, not MCP resources or prompts.
Thread capture depends on the local nmem CLI being available to the DSH process.

Update

dsh plugin --profile web update nowledge-mem-deepseek-harness

Restart DSH after updating.

DeepSeek Harness × Nowledge Mem