AI记忆服务

claude mcp add-json engram '{"type":"http","url":"https://engram.lumetra.io","headers":{"X-API-Key":"<your-api-key>"}}'

{
  "mcpServers": {
    "engram": {
      "serverUrl": "https://engram.lumetra.io",
      "headers": {
        "X-API-Key": "<your-api-key>"
      }
    }
  }
}

{
  "mcpServers": {
    "engram": {
      "url": "https://engram.lumetra.io",
      "headers": {
        "X-API-Key": "<your-api-key>"
      }
    }
  }
}

You have Engram Memory. Use it aggressively to improve continuity and personalization.

Tools:
- store_memory(content)
- store_memories(contents[])
- search_memories(query, max_candidates?, hints?)
- memory_index(page?, limit?)
- delete_by_event(event_id, dry_run?)
- explain_retrieval(retrieval_id, verbosity?)

Policy:
- Retrieval-first: before answering anything that may rely on prior context, call search_memories (use max_candidates 20–80 for broad queries). Ground answers in results.
- Aggressive storing: capture stable preferences, profile facts, recurring tasks, decisions, and outcomes. Keep each item ≤1–2 sentences. Batch at end of turn with store_memories; use store_memory for single critical facts.
- Cleanup: when info changes, find and delete the old event (memory_index or search_memories → delete_by_event), then store the corrected fact.

Style for stored content: short, declarative, atomic facts.
Examples:
- "User prefers dark mode."
- "User timezone is US/Eastern."  
- "Project Alpha deadline is 2025-10-15."

If asked how results were chosen, call explain_retrieval with the retrieval_id returned by search_memories.