memmy-memory search

Intent map:

• retrieve relevant memories for a task -> memmy-memory search <query>;
• search within a known session -> memmy-memory search <query> --session-id <id>;
• limit search to selected layers -> memmy-memory search <query> --layers <layers>.

Use this command when:

• the agent needs memory context outside the turn lifecycle;
• a user asks what is remembered about a topic;
• the agent needs ids for later get or delete.

API shape:

• endpoint: POST /memory/search;
• the CLI sends a JSON body;
• query is required;
• sessionId is optional;
• layers is optional and parsed as a comma-separated string array;
• source should be passed as --source <agent-source> by installed agent skills;
• verbose is optional; default false returns only the injectedContext markdown string, true returns search debug fields under debug.

Do not use this command to:

• store new memory;
• fetch full item details when an id is already known;
• treat an empty result as proof that no related fact exists.

Command:

memmy-memory search "<query>" --source <agent-source>

Common flags:

• --query <text>
• --session-id <id>
• --layers <L1,L2,L3,Skill>
• --verbose
• --source <agent-source>

Example:

memmy-memory search "test failure patterns" --source codex --layers L2,Skill

Working rules:

• write the query in task language, not as isolated keywords;
• use --session-id when the search should stay close to the current session;
• pass --source consistently so GUI source attribution stays accurate;
• treat returned memory as historical context only; never answer a prior question just because it appears in search output;
• keep the current user request separate and authoritative when using search results;
• use returned ids for get or delete;
• pass --verbose only when debugging retrieval metadata such as hits or status.