Memory and Knowledge Tools (Apprentice MCP)

These built-in tools give an agent durable context across runs. They cover long-term memory, knowledge base search, a private key-value store, and conversation summaries.

All tools on this page are gated by the Persistent Memory capability. If that capability is off, none of them are available to the agent.

What These Tools Are For

• Memory: remember facts and recall them later, including a graph view of related memories.
• Knowledge base: search files you uploaded to the agent.
• Store: a private key-value store the agent can use as scratch state.
• Summaries: search and read condensed summaries of past work.

Memory and knowledge live in local Apprentice data and are scoped to the agent. When an external provider needs them to answer, the relevant context is included in the model call.

Memory Tools

• memorize: save an important fact to long-term memory.
• remember: store a memory entry, optionally with structure or tags.
• recall_memories: retrieve memories relevant to the current context.
• recall: fetch a specific memory or set of memories.
• search_memories: search stored memories by query.
• search_graph: traverse related memories through the memory graph.
• forget: remove a memory the agent no longer needs.

Use memory for facts that should survive across sessions, such as user preferences, project conventions, or decisions.

Knowledge Base Tools

• get_kb: read knowledge base content available to the agent.
• search_kb: search the agent's uploaded knowledge files.

Use the knowledge base for reference material you upload, such as product docs, policies, or specifications. Knowledge files are associated with the agent that uses them.

Store Tools

• store_put: write a value under a key.
• store_get: read a value by key.
• store_has: check whether a key exists.
• store_delete: delete a key.
• store_list: list keys, optionally within a namespace.
• store_namespaces: list available namespaces.
• store_clear: clear keys in a namespace.

Use the store for structured working state the agent manages itself, such as counters, cursors, or small lookup tables.

Summary Tools

• get_summaries: list available summaries.
• search_summaries: search summaries by query.
• get_summary_chunk: read a specific chunk of a summary.

Summaries let an agent recover the gist of long past conversations without rereading everything.

Designing Around Memory

Good patterns:

• Tell the agent in its instructions when to call memorize and what is worth remembering.
• Prefer search_kb for uploaded reference material and recall_memories for learned facts.
• Use the store for machine state, not for facts a human would want to read.

Next Step

Continue with Collaboration and Messaging Tools to let the agent work with channels, chats, topics, and other agents.