Persistent memory for AI agents — REST API, MCP, OAuth, CLI, dashboard. Semantic search, knowledge graph, automatic consolidation. One self-hosted service, every transport.
Native Milvus batch update (1 round-trip), SSE reconnect replay, and growing-segment dedup fix. ~1,972 tests.
Milvus backend now overrides update_memory and update_memories_batch. The batch variant performs a single fetch + batch embedding + single upsert — one round-trip instead of N, dramatically reducing latency for bulk update workloads.
A bounded ring buffer lets SSE clients resume missed events after a transient disconnect using the standard Last-Event-ID header. Buffer size is configurable via MCP_SSE_REPLAY_BUFFER_SIZE (default 1000). Replay outcome is surfaced in the connection_established welcome event.
Replaces ANN search() with brute-force query(consistency_level="Strong") + client-side cosine similarity in semantic deduplication, fixing Milvus Lite growing-segment visibility where freshly inserted records were silently missed.
Claude.ai connects to your memory server through Streamable HTTP with OAuth 2.1 authentication.
Battle-tested with comprehensive testing and optimized for performance.
Install from PyPI and connect your AI agents to persistent memory.