Profile-based long-term memory backend that extracts and consolidates conversation turns into structured user profile slots (topics, preferences, facts) stored in PostgreSQL. Incoming data is queued in a Redis buffer and batch-processed by an LLM into the profile, decoupling write cost from request latency. At read time the pre-compiled profile is returned via SQL lookup (sub-100ms); an event timeline of timestamped entries sits alongside the profile and supports full-text and tag-based search. Self-hostable via Docker Compose (FastAPI + Postgres + Redis); managed cloud also available.
- Storage
- PostgreSQL (structured user profile slots + event timeline) with a Redis buffer for incoming unprocessed data; cloud-managed or self-hosted via Docker Compose
- Retrieval
- SQL lookup of pre-compiled profile slots at sub-100ms latency; no per-query LLM call at retrieval time; optional embedding search over the event timeline for semantic context; context assembled into a prompt prefix on demand
- Self-host
- Self-host: moderate
- License
- Apache-2.0
- Pricing
- Free tier (managed cloud); pay-as-you-go tiers for power users and high-volume teams; self-hosted OSS is free · Free + paid
- GitHub stars
- 2,800
- Last release
- 2026-01-11
- Last commit
- 2026-01-11
- First catalogued
- 2026-06-28
Strengths
- Sub-100ms profile retrieval: pre-compiled slots are SQL-read at inference time, eliminating per-query LLM calls
- Batch buffer amortizes LLM extraction cost; self-reported ~5x cheaper and ~5x faster than mem0 at equivalent quality
- Multi-language SDK surface: Python, Node.js/TypeScript, and Go SDKs plus a REST API and a stdio MCP server
- Event timeline alongside the structured profile enables temporal queries, full-text search, and tag-based filtering over past interactions
- Controllable profile schema: operators define slot types and token budgets, giving predictable cost and output shape
Watch out
- Profile-centric design targets user-preference and fact memory; episodic or agent-task memory is outside the primary scope
- No formal GitHub Releases — version history is git tags only; latest tag v0.0.42 on 2026-01-11 (~5.5 months without a commit as of Jun 2026); verify maintenance status before adopting
- Performance claims (5x cheaper/faster vs mem0, LOCOMO SOTA) are self-reported without published backbone/embedder configs; no independent reproductions available
Best for
- Chat applications and AI companions requiring persistent, structured user profiles (preferences, topics, facts) at sub-100ms retrieval latency
- Multi-session educational tutors, customer-service bots, or voice agents (LiveKit adapter available) where long-term user history must be structured and queryable
- Teams wanting a self-hostable or cloud-managed memory REST backend with multi-language SDKs and an MCP server for IDE or agent integration
How it integrates
Benchmark results
No sourced results yet.
Sources
- https://www.memobase.io/ (vendor)
- https://github.com/memodb-io/memobase (vendor)
- https://github.com/memodb-io/memobase/tags (vendor)
- https://docs.memobase.io/llms.txt (vendor)
- https://docs.memobase.io/cost (vendor)
- https://www.pulsemcp.com/servers/memobase (third-party)
Last verified 2026-06-29 · updated by refresh-framework-cards