Introduction

DSH’s capability expansion is usually done via plugins. A common problem is context discontinuity between sessions: preferences confirmed in the current session, project conventions, and queried data may not be usable in the next session.

interest-memory addresses this by providing a local long-term memory service: it organizes and writes data when a session ends and recalls relevant entries when a session begins. DeepSeek Harness can integrate this service via the DSH bridge plugin.

What is it

Maintained by djasdh, interest-memory belongs to the category of memory capabilities. It breaks down cross-session long-term memory into a local Go service that uses SQLite-vec for vector indexing and provides a verification process.

For DeepSeek Harness, the integration form is a Cordis plugin; the installation target is @djasdh/interest-memory-dsh-bridge. The service uses a single binary and an SQLite file as the local knowledge base and also supports multiple agent frameworks sharing the same service.

Core Capabilities

Below are the verified capabilities.

Session Integration

  • DeepSeek Harness connects via the Cordis plugin.
  • Recall injection is performed at the start of a session.
  • Transcript push is performed at the end of a session.
  • Provides memory_search / memory_logs consumer tools.

Local Storage & Retrieval

  • Uses SQLite-vec vector indexing and provides FTS (Full-Text Search) fallback.
  • Uses a single binary + SQLite file as the local knowledge base.
  • Entries carry evidence, such as web URLs, turns, and queries.

Memory Governance

  • Uses a 3-stage verification: check / claims / contradictions.
  • Semantic similar interest points are automatically merged or associated.
  • Old entries are marked as superseded / archived, and a replacement chain is retained.
  • LLM writes to wiki pages and establishes semantic edges.

Query & Visualization

  • Search results carry outlinks / backlinks and support graph walk.
  • Provides REST API: sessions, recall, search, logs, interest-points, wiki pages, fork, jobs, stats, graph.

Installation & Enablement

First install the local service, then install the DSH bridge.

Install Service

Use the installation script:

curl -fsSL https://raw.githubusercontent.com/djasdh/interest-memory/main/scripts/install.sh | bash

This command executes the installation script.

Configure LLM

If you need to use LLM for writing, verification, or embedding, you can run:

curl -fsSL https://raw.githubusercontent.com/djasdh/interest-memory/main/scripts/install_llm.py | python3 - --provider <provider>

Replace <provider> here with the actual provider. The documentation indicates that LLM / embedding can point to local Ollama / vLLM for fully offline scenarios.

Install DeepSeek Harness bridge

Install the bridge in DSH:

dsh plugin --profile web add @djasdh/interest-memory-dsh-bridge

This command installs the DSH bridge, used to connect interest-memory to DeepSeek Harness.

Typical Usage

The following interfaces use {agent} as the agent path segment.

Push Session End Content

POST /api/v1/{agent}/sessions

Used to push the session-end transcript when a session ends.

Recall Injection

GET /api/v1/{agent}/recall?query=&after=&before=&days=

Used to recall relevant memories at the start of a session.

Query Full Content

GET /api/v1/{agent}/search?query= or ?id=&top_k=

Used to query full content and graph information.

View Change Log

GET /api/v1/{agent}/logs?limit=&offset=

Used to view the change log.

Scenarios & Considerations

Suitable for users who want to consolidate cross-session information within DSH while retaining local storage, queryability, and an audit trail. It can also serve as a local memory service shared by multiple agent frameworks.

Please note:

  • The service will not block sessions if it is unavailable.
  • session_transcripts saves the complete original text; if you are concerned about disk growth, you need to trim it externally.
  • fork.max_concurrency / verify.max_concurrency are used to limit peak memory usage.
  • Kanban boards listed in interestmemory.kanban_exclude will be discarded.
  • Shared results will be tagged with a source: recall lines use [from: <agent>], search / get use result.agent.
  • The plugin runs with the current dsh process permissions. You should check the source code and license before installing; currently, verified materials have not confirmed the license.
  • The plugin directory is an independent site and the directory page should not be considered an official app store for DeepSeek or Huxuan.

Links

  • GitHub: https://github.com/djasdh/interest-memory
  • Plugin Directory: https://www.skillhub.cn/plugins/djasdh/interest-memory (This link comes from plugin clues and was not directly confirmed in the materials scraped this time)