Preface

The philosophy of DSH is “Everything is a plugin.” For agent developers, the model context window is limited. Sessions, notes, and insights accumulated across sessions are difficult to continuously utilize without external storage.

dsh-llmwiki solves this type of problem: it uses a local Markdown vault as DeepSeek Harness’s long-term memory and injects relevant Wiki knowledge into the same model request.

It is maintained by chancelu and is licensed under MIT. The DSH community directory is a separate site, not an official app store; it has no official affiliation with DeepSeek / High-Flyer.

What is this

dsh-llmwiki is a DSH plugin designed as DeepSeek Harness’s local Markdown long-term memory. Ported from llmwiki, it serves to put historical sessions, notes, and persistent insights into a local Markdown vault and recall them in subsequent model requests.

Core Capabilities

The confirmed capabilities are introduced below:

  • Inject relevant Wiki knowledge into the same model request.
  • Explain memory capabilities to the model.
  • Provide the memory_search tool to recall historical sessions/notes.
  • Provide the memory_save tool to save persistent insights.
  • Automatically capture to chronicle/daily/YYYY-MM-DD.md every round.
  • Use three retrieval strategies: keyword, wikilink graph, and temporal, and fuse results using Reciprocal Rank Fusion (RRF).
  • Assemble context within a token budget.
  • Use LRU + TTL caching.
  • No runtime dependencies other than Node.js.

Installation and Enablement

First, confirm the environment:

# Node.js ≥ 22
# Available dsh CLI, for example:
npm install -g @deepseek-ai/dsh
# pnpm available in PATH

Then install the plugin:

dsh plugin --profile web add dsh-llmwiki

After installation, first view the config layer:

dsh --profile web --dump-config

This step is used to view the config layer containing # == dsh-llmwiki.

Next, start:

dsh web

This step starts and loads the plugin.

Configuration

Zero configuration by default; the vault is located at:

~/llmwiki-vault

If you need to override configuration, add a config line for id: llmwiki in the profile’s cordis.patch.yml, and fully restate the items to be retained.

Note: Config patches replace the entire config line, so all keys to be retained need to be restated.

Typical Usage

First do the installation, then confirm the config layer, and then start:

  1. Run the installation command.
  2. Run dsh --profile web --dump-config to view the plugin config layer.
  3. Run dsh web.
  4. Call memory_search in a session to recall historical sessions/notes.
  5. Call memory_save to save persistent insights.
  6. Check chronicle/daily/YYYY-MM-DD.md to confirm that the automatically captured content has been written.
  7. Open the vault with Obsidian and organize Layer-3 notes using [[wikilinks]]; the wikilink graph retrieval strategy will use these links.

Applicable Scenarios and Notes

Suitable for the following scenarios:

  • You want DeepSeek Harness’s long-term memory to reside in a local Markdown vault.
  • You are already using plugins in your DSH workflow and want to integrate memory/long-term memory capabilities.
  • You want to organize notes using local Markdown tools.

Notes before use:

  • The plugin reads and writes to the local vault and participates in model context assembly.
  • The plugin runs with the permissions of the current dsh process. You should check the source code and license before installing; the license is MIT.
  • The ripgrep / SQLite FTS engine and the LLM-driven curate pipeline have not been ported yet.
  • Requires Node.js ≥ 22, an available dsh CLI, and pnpm in PATH.

Conclusion

The value of dsh-llmwiki is to separate long-term memory from the model context and place it in a local Markdown vault, allowing DSH to continue using this knowledge in subsequent rounds through retrieval, injection, saving, and automatic capture.

GitHub Link:

https://github.com/chancelu/dsh-llmwiki

Directory page: No link provided in verified information.