Skip to main content
Memory protocol methods let clients inspect and manage durable agent memory directly. The agent can also use model-visible memory tools during agent-mode turns, but those tools are separate from the client JSON-RPC protocol. All memory methods are served by the gateway and routed through MemoryService. Clients should treat the service response as authoritative even when the underlying capsule backend has stale search data. This page is about durable memory, not thread episodic context. Durable memory is where Pioneer stores long-lived facts and decisions. Thread episodic context is the searchable conversation-history layer; it is controlled through gateway settings and used by the agent recall path, but it is not exposed as the memory/* JSON-RPC API.

Methods

Candidate review methods exist even though the default product path currently rejects/suppresses middle-confidence candidates instead of surfacing a review queue to users.

Notifications

Clients should listen for these events if they render memory management screens or cached memory lists. Thread episodic vector indexing has a separate gateway notification, gateway/thread_episodic/vector_refill/status_changed. It reports workspace-scoped refill progress such as running, complete, or failed. Clients that show semantic-search readiness should use this event together with settings and the memory/embedding model status; it is not a durable-memory mutation event.

Core dtos

Generated schemas are exported under schemas/ in the repository. The protocol source is crates/protocol/src/memory.rs.

Durable memory vs thread context

The protocol intentionally keeps these surfaces separate: Do not use memory/remember to store every thread message. That would pollute durable memory and make forgetting/ranking much harder. If a client wants to display transcript history, use thread and turn APIs. If the agent needs older discussion as context, the thread episodic layer handles that through bounded recall.

Candidate statuses

Candidate statuses include active review states and dormant states: The default runtime does not currently use pending_silent, ask_on_use, or needs_review for user-facing review.

Semantic write contract

Semantic writes are used by internal extraction and future advanced clients. They include:
  • semantic fields: intent, explicitness, category, subject, attribute, scope hint, durability, sensitivity, certainty;
  • normalized content and optional value;
  • evidence with source thread/turn/item/ref and quote/span;
  • provenance actor;
  • confidence and importance;
  • disposition such as route_to_candidate_policy.
Even when a caller provides semantic fields, the service still owns canonical key generation, duplicate detection, scoring, and final state.