Your memory, over MCP.

Saves a new memory to the user's account, optionally into a specific bucket. Duplicates are detected and ignored.

When to use it: Whenever something worth remembering comes up: a preference, a decision, project details, personal context. Assistants should err on the side of saving and ask which bucket to use when it is unclear.

Hybrid semantic and keyword search over saved memories, ranked by relevance.

When to use it: When looking for specific saved information: a name, a preference, a decision. The best first call when the user references something they have told an AI before.

Returns recent (or all) memories plus the full bucket list in one call. Accepts an optional search query.

When to use it: At the start of a conversation to load context, or when the assistant needs both memories and the bucket structure at once.

Edits a single memory's text or bucket, or moves up to 100 memories to another bucket in one call. Buckets can be referenced by ID or by name; a named bucket that does not exist is created.

When to use it: When the user corrects a saved fact, or wants memories reorganized. Bulk moves are all-or-nothing: one unresolvable ID rejects the whole request.

Lists the user's memory buckets.

When to use it: Before storing into or searching a specific bucket, or when the user asks how their memories are organized.

Creates a new bucket.

When to use it: When the user wants a new folder for a topic, project, or area of life. Ask for a name if none is given.

Lists a bucket's Smart Memory categories. Each category carries a ~200-word summary and pointers to what its full memories contain, plus the bucket's 30 most recent memories.

When to use it: To get a token-efficient overview of a large bucket. Read the summaries first, then load only the relevant category with list_category_memories.

Returns the full memories inside one Smart Memory category.

When to use it: After list_bucket_categories shows a category is relevant to the current conversation.

Searches past conversations and returns synthesized summaries with source citations, not raw logs. Supports up to 15 parallel queries, date-range filters, a token budget per query, and a slower quality mode for hard questions.

When to use it: When the user asks about past decisions, projects, people, or anything their conversation history can answer. Use one query for simple lookups; for multifaceted questions, send parallel queries from different angles (timeline, people, decisions, outcomes).

Returns an AI-generated summary of one conversation. Conversations under 5,000 tokens return the full transcript instead.

When to use it: To dive into a specific conversation surfaced by recall_chat_history. Takes the conversationId from that tool's sources array.

Returns the complete transcript of one conversation, unsummarized.

When to use it: When the exact wording matters or a summary is not enough. Same conversationId as get_conversation_summary.

Exports the full normalized transcript of one conversation as a JSON file and returns a temporary download URL (expires after 15 minutes, no auth) that the agent fetches to get the structured messages.

When to use it: When the agent needs the whole conversation as structured data: too long to read inline, or destined for file-based analysis or scripting. Read short transcripts with get_full_conversation instead.

Returns an AI-generated overview of the user derived from their whole chat history.

When to use it: For a broad picture of who the user is and what they work on, before any specific lookup.

Semantic search over the user's uploaded documents. Returns relevant passages with the source file and page numbers.

When to use it: When the user references their MemoryPlugin documents, or asks about "my files" and no file is attached to the current conversation.