Storage APIs for AI agents

Letta — create a stateful agent with self-managed memory blocks. Returns an agent `id` to drive with `letta:message`. Optional: name, model, memory_blocks.

Letta — send a message to an agent; the agent thinks and self-edits its memory. Pass `agent_id` (from agent-create) and `input` (the user message). Reply is…

Mem0 — store conversation turns as long-term memory. Pass `messages` ([{role,content}]) and a `user_id`; Mem0 auto-extracts facts. Returns a queued event id.

Mem0 — semantic search over a user’s stored memories. Pass `query` and `user_id`; returns ranked memories with scores.

Zep — add messages to a thread; Zep ingests them into the user’s knowledge graph. Pass `thread_id` and `messages` ([{role,content,name?}]). Set…

Zep — fetch the token-efficient summarized context block for a thread (drop it into your LLM prompt). Pass `thread_id`.

Zep — search a user’s knowledge graph for specific facts (vs the summarized context). Pass `query` and `user_id`. Optional scope (edges|nodes|episodes) and…

Zep — open a thread (conversation container) for a user. Pass `thread_id` and `user_id`. Facts ingested in one thread are available across all the user’s…

Zep — create a user (prerequisite before threads/messages). Pass a stable `user_id`. Optional: email, first_name, last_name, metadata.