AI Agent Hub
Back to plugins
🧠

dsh-gentle-engram

Memory Updated 2026.08.27

Run the following command in DeepSeek Harness:

dsh plugin install eehcx/dsh-gentle-engram

Paste the following prompt into your AI chat to install this plugin:

Install dsh-gentle-engram in DeepSeek Harness; the source is at https://github.com/eehcx/dsh-gentle-engram and requires Engram to be installed and available as engram in PATH.

About this plugin

DeepSeek Harness lacks a built-in, stable memory lifecycle, and Engram is the missing piece. dsh-gentle-engram is a TypeScript-based Cordis plugin that bridges Engram's MCP stdio tools into DSH sessions, hooking into session start, inbound prompts, tool results, turn stopping, and agent disposal so that relevant memory context is seeded before every model step while prompts and filtered passive tool learnings are persisted asynchronously.

On the reliability front, per-session writes are serialized and all pending writes are flushed at turn boundaries to prevent loss or reordering; when an agent is disposed, the plugin closes the Engram session and records a structured summary. Security-wise, it never touches Engram's SQLite database directly and filters out likely secrets before any data leaves the process.

It is built for DeepSeek Harness developers who want persistent memory with minimal setup: a single dsh install command gives you a quiet background that handles context injection and learning capture, without manual MCP row management or hand-rolled lifecycle hooks.

Use Cases

  • Auto-seed Engram memory context into every model step of a DSH session
  • Serialize and flush pending writes at turn boundaries for reliable persistence
  • Gracefully close Engram sessions and record a structured summary on agent disposal

Best For

  • Developers running DeepSeek Harness with Engram for persistent memory
  • Users who want a hands-off memory experience without manual MCP setup
  • Integration engineers who care about write reliability and secret isolation