Peon Memory — User Guide
What is it?
Peon Memory lets you teach the AI workspace-specific rules and preferences that persist across sessions. Once stored, every conversation automatically includes your guidelines so the AI respects them without repetition.
Available since v1.7.5.
How to use it
Trigger phrases
Use any of these keywords in a chat message to save something:
| Phrase | Example |
|---|---|
remember this | "Remember this: always format Java files with 4-space indentation" |
always do | "Always do a clean build before running tests" |
never do | "Never do inline comments in production code" |
The AI extracts the guideline and stores it automatically — no manual configuration needed.
Managing memory
You can also ask the AI directly to manage stored guidelines:
- View: "Show me my workspace memory" or "What do you remember?"
- Remove: "Remove guideline #3" (refers to the numbered list shown by the AI)
- Edit: "Change guideline #2 to ..."
- Reset: "Clear all memory" (requires explicit confirmation)
How it works
Chat history persistence
Peon also persists the current chat history for Peon-Dev, Peon-Plan, Jon (Peon-PO), and custom agents, so their conversation context survives Eclipse/plugin restarts.
History lives in the workspace metadata state — one file per agent at <workspace>/.metadata/.plugins/org.sterl.llmpeon/state/<agent>-history.jsonl. Because it is workspace-scoped, each workspace keeps its own history and multiple instances of the same project never collide. The shared config directory ~/.peon now holds only configuration (Agents, Skills, Commands). On first start a one-time, automatic migration moves any existing ~/.peon/state history into the workspace and removes the now-empty directory. If a file already exists at the target it is skipped — the existing file wins, the legacy copy is kept — and the skip is logged; the migration never overwrites data.
The Clear button deletes only the active agent's chat history and queued messages. It does not delete Peon Memory guidelines; use the memory reset command for that.
Guidelines are injected as a Memory: block at the start of every conversation. The AI sees them before processing your request and follows them whenever they apply.
Storage is workspace-scoped — each Eclipse workspace maintains its own independent set of guidelines. A maximum of 500 entries is supported; when full, the oldest guideline is automatically removed.