NeoSapien

Export specific memories to JSON or CSV format.

export_memories

Get the latest N memories for a specific person. Convenience tool — not a paginator. Optimized for person-centric queries like "What did <name> say recently?". Use search_owners to find valid owner names first. For full pagination across all memories for a person, use `search_memories` with owner_name and page=1, page=2, ... instead.

get_latest_memories_by_owner_name

Fetch the FULL transcript for a memory. Use this ONLY when you strictly need the raw conversation text. Returns the list of transcript segments.

get_memory_transcript

Retrieve full details for a SINGLE memory ID.

get_memory_by_id

Get reminders (action items) extracted from the user's memories. Each reminder links back to the memories it was extracted from via `source_memory_ids`. Results are paginated.

get_reminders

Get ALL memories indiscriminately. Use this only for "Show me everything" requests or browsing history without filters. Results are paginated.

list_all_memories

Get specific memories by their exact IDs. Use this implementation mostly for internal pagination or if you already have a list of definitive Memory IDs from another source and just need to fetch their content. For searching, use `search_memories` instead.

list_filtered_memories

PRIMARY TOOL for searching memories. Combines robust filtering with content retrieval. Use this for 90% of memory search tasks. Capabilities: - Full-text search (title, summary, MOM) - Filter by Owner Name (resolves an owner name from search_owners to a user ID) - Filter by Tags, Entities, Domains, Topics, and Date Ranges - Returns light memory objects (summary, metadata) WITHOUT transcripts/MOM by default to save context - Supports pagination — use page=2, page=3, etc. to retrieve all results PAGINATION GUIDANCE: When total_found > returned, there are more results. Use total_pages from the response to know how many pages exist, then call this tool again with page=2, page=3, ... until all pages are fetched. ALWAYS paginate through ALL pages before responding — this applies to ANY request involving a time range, topic, person, tag, or keyword. Never summarise, analyse, or answer based on a partial result set. If the user asks "what happened this week", "everything about X", "all meetings with Y", or any similar broad query, fetch every page first, then respond.

search_memories

Get aggregated STATISTICS and METADATA (counts, top tags, etc.) ONLY. WARNING: This tool DOES NOT return memory content/text. Use this ONLY when you need to know "How many memories?" or "What are the top topics?". Use `search_memories` if you need to read the actual memories.

search_memories_metadata

Search for memory owners (people) by name. Use this when you are unsure about the spelling of a name or want to list available people.

search_owners