Store a new memory. Requires at least one of user_id, agent_id, or run_id.

    Response contract (never a bare list):
    - {"status": "queued", "task_id", "submitted_at", "queue_depth", "estimated_wait_s"}
      — infer=true path; extraction runs in background, poll memory_task_status.
    - {"status": "stored", "memory_ids": [...], "results": [...]}
      — synchronous path; empty memory_ids carries "reason": "no_new_facts".
    - {"error": ...} — failure.
    

Ingest a PDF or image asynchronously: extract text per page (digital PDF via poppler; scanned pages and images via a local vision model), chunk, and extract memorable facts with document/page provenance.

    Returns immediately with {"status": "queued", task_id, pages,
    chunks_estimate, estimated_wait_s} — a large document takes many
    minutes; poll memory_task_status(task_id) for chunks_done progress and
    the final memory_ids. Re-submitting the same file returns
    {"status": "already_ingested"} unless force=true. There is NO
    synchronous fallback: if the queue is unavailable the call errors.
    

Semantic search across existing memories.

Page through memories using filters instead of search.

Fetch a single memory by its ID.

Full change timeline of a memory: ADD, UPDATE (old vs new text), SUPERSEDED (which fact replaced it) and DELETE events, oldest first.

Status of an asynchronous add_memory task.

    States: pending | processing | done | failed_retryable | dead.
    When done, ``result.memory_ids`` lists the memories created/updated
    (fetch them with get_memory) and ``result.events`` includes any
    SUPERSEDED markings. ``last_error`` explains failed/dead tasks.
    

Health of the async ingest queue: depth (jobs waiting or running), per-status counts, age of the oldest pending job, estimated drain time, and whether the background worker is alive.

Overwrite an existing memory's text. Re-embeds and re-indexes.

Delete a single memory.

Bulk-delete all memories in the given scope. Requires at least one filter.

    NEVER calls memory.delete_all() — uses safe bulk-delete instead.
    

List which users/agents/runs currently hold memories.

Uses Qdrant Facet API (v1.12+) for server-side aggregation, with scroll+dedupe fallback for older versions.

Delete an entity and cascade-delete all its memories.

    Functionally equivalent to delete_all_memories in self-hosted mode.
    

Search entities by name/id substring matching in Neo4j knowledge graph.

Get all relationships for a specific entity (bidirectional).