nowledge-mem-deepseek-harness: A Unified Memory Layer DSH plugin for DeepSeek Harness

nowledge-co/nowledge-mem-deepseek-harness

Memory & ContextVerified
Listed on 2026-08-20
Page last updated 2026-08-20

A durable memory layer for DeepSeek Harness with startup context, prompt-time recall, and MCP memory tools.

Let DeepSeek Harness share the same durable memory system as other AI tools and agents, solving the lack of long-term memory in sessions.

install
dsh plugin --profile web add github:nowledge-co/nowledge-mem-deepseek-harness
Category
Memory & Context
Platform
DSH-Plugin
Author
nowledge-co
Distribution
Plugin

nowledge-mem-deepseek-harness Key Features

Startup context injectionPrompt-time recallMCP memory toolsTurn-end thread capture

nowledge-mem-deepseek-harness Repository

nowledge-co
Publisher
nowledge-co
Repository
nowledge-co/nowledge-mem-deepseek-harness
5
Stars
1
Forks
0
Watchers
0
Open issues
LanguageJavaScript
License
Website
Repo last push2026-08-17 10:12:52
Repo created2026-08-13 21:47:02
Topics
deepseek-harnessdsh-pluginknowledge-graphmcpmemorynowledge-memontology
Default branchmain

nowledge-mem-deepseek-harness Repo Summary

Nowledge Mem is a memory and context plugin (DSH plugin) for DeepSeek Harness, maintained by nowledge-co and written in JavaScript. It provides a unified persistent memory layer for DSH, enabling DSH to share the same memory system as other AI tools and agents. The plugin solves the problem of DSH sessions lacking long-term memory and cross-tool context reuse. Its core capabilities include startup context injection, prompt-time recall, MCP memory tools, and turn-end thread capture. The plugin injects the Nowledge Mem Context Bundle once per DSH session via agent/pre-step, and when the user asks to continue, remember, review previous work, ship a release, debug a regression, or discuss connectors/plugins, it calls nmem --json m search "<prompt>" -n 8 for memory recall. By default, it connects to the local Mem MCP endpoint http://127.0.0.1:14242/mcp/, and supports connecting to Nowledge Cloud or another remote Mem via environment variables NMEM_MCP_URL and NMEM_API_KEY. The plugin registers tools under the nowledge_mem namespace through DSH's MCP bridge, and after each completed turn, it imports the real DSH surface transcript using nmem t import --source deepseek-harness, stamping imports with NMEM_IMPORT_ORIGIN=deepseek-harness.

What core features does the Nowledge Mem DSH plugin support?

Nowledge Mem offers four core capabilities: Startup Context, Prompt-Time Recall, MCP Tools, and Thread Capture. Startup Context injects a user message marked as form: "snapshot" at session start, reminding the model that it is cross-tool context, not an instruction override. Prompt-Time Recall performs memory searches for specific prompt types (e.g., continue, release, regression debugging), returning a bounded recall message with memory titles, source hints, scores, and content. MCP Tools are registered through DSH's MCP bridge, with tool names like mcp__nowledge_mem__.... Thread Capture serializes user, assistant, and tool-result events after each completed turn, skips its own injected context messages, and imports the transcript into Mem as source=deepseek-harness.

How to install the Nowledge Mem DSH plugin?

To install Nowledge Mem, ensure the nmem CLI is on PATH, then add the plugin and start the DSH web interface:

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

For a local checkout of nowledge-co/community, run from the repository root:

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

After installation, verify the nmem CLI is available:

bash
nmem status
nmem config mcp show --host deepseek-harness

How to configure the Nowledge Mem DSH plugin?

Configuration can be done via row config fields in a later cordis.patch.yml override, such as cliPath, sourceApp, importOrigin, contextOnSessionStart, recallOnPrompt, syncOnTurnEnd, recallLimit, spaceId, and agentId. Ambient variables also work, including NMEM_SPACE, NMEM_AGENT_ID, NMEM_HOST_AGENT_ID, NMEM_MCP_URL or NOWLEDGE_MEM_MCP_URL, and NMEM_API_KEY. By default, it connects to the local Mem MCP endpoint; for Nowledge Cloud or another remote Mem, set NMEM_MCP_URL and NMEM_API_KEY.

What are the known limitations of the Nowledge Mem DSH plugin?

DeepSeek Harness is in developer preview, so public package APIs may change. The plugin does not backfill historical DSH sessions; it only captures new sessions while the plugin is active. The DSH MCP bridge exposes tools only, so Nowledge Mem resources and prompts are not shown through MCP today. Turn capture depends on the local nmem CLI being available in the DSH process environment.

What is the community position of the Nowledge Mem DSH plugin?

This repository is the canonical standalone plugin package, mirrored in nowledge-co/community for the Nowledge Mem connector index. It is maintained by nowledge-co, last updated on 2026-08-16, written in JavaScript, with topics including deepseek-harness, dsh-plugin, knowledge-graph, mcp, memory, nowledge-mem, and ontology.

View more DSH plugins at dsh-plugin.org

DSH-Plugin FAQ