memory-search（记忆搜索）

Memory Search

You have two tools for recalling information from your memory files. Use them.

Tools

memory_search

Semantic vector search across your indexed memory files (MEMORY.md, memory/*.md, and session transcripts).

Parameters:

Example calls:

{ "query": "what projects is the human working on" }
{ "query": "preferences about code style", "maxResults": 3 }
{ "query": "important dates birthdays deadlines", "maxResults": 10, "minScore": 0.3 }

Returns: Array of results, each with:

• snippet — the matching text chunk
• path — relative file path (e.g. MEMORY.md, memory/2026-02-07.md)
• startLine / endLine — line range in the source file
• score — relevance score
• citation — formatted source reference (in direct chats)

memory_get

Read a specific section of a memory file by path and line range. Use this after memory_search to pull more context around a result.

Parameters:

Example calls:

{ "path": "MEMORY.md" }
{ "path": "memory/2026-02-07.md", "from": 15, "lines": 30 }

When to Use Memory Search

Always search before answering about:

• Prior conversations or decisions
• The human's preferences, habits, or opinions
• Dates, deadlines, birthdays, events
• Project status or history
• Anything the human said "remember this" about
• Todos, action items, or commitments
• People, names, relationships

The pattern is:

1. Receive a question that might involve past context
2. Call memory_search with a relevant query
3. Review the results
4. If a snippet looks promising but needs more context, call memory_get with the path and line range
5. Answer using what you found (cite sources in direct chats)

When NOT to Use

• Purely factual questions with no personal context ("what is Python?")
• The human explicitly gives you all the context you need in the message
• You just searched and the results are still in your context

Tips

• Be specific in queries. "birthday" works better than "important information about the human."
• Search multiple angles. If one query returns nothing useful, try rephrasing. "project deadlines" and "what's due soon" might return different results.
• Don't over-fetch. Start with default maxResults. Only increase if you need more coverage.
• Use memory_get sparingly. The search snippets are usually enough. Only pull full sections when you need surrounding context.
• Say when you checked. If you searched and found nothing, tell the human: "I checked my memory and didn't find anything about that." Don't silently guess.

What Gets Indexed

Your memory search covers:

• MEMORY.md — your curated long-term memory
• memory/*.md — daily notes and raw logs
• Session transcripts (if enabled)

These files are automatically indexed. You don't need to trigger indexing — just write to the files and the system handles the rest.

Do NOT

• Do NOT try to run shell commands like cat or ls to read memory files. Use memory_search and memory_get.
• Do NOT try to configure or debug the search system. That's operator config, not your job.
• Do NOT assume memory is empty without searching first. The index may have content even if the memory/ directory looks sparse.