Memobase is a Python-based user profile memory system for long-term memory in LLM apps, offering a Python SDK, APIs, and Docker-ready architecture with Postgres and Redis backing. It targets memory, event timelines, and prompt integration with batch processing and low latency.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
Memobase is a user profile-based memory system designed to bring long-term user memory to LLM applications. It emphasizes memory for the user rather than the agent, with time-aware memory through user events and a configurable profile system. It provides integrations via API and SDKs across Python, Node, and Go, and is built to batch-process chats per user to optimize latency and cost.
How it works
Memobase maintains a per-user profile and event timeline. Blobs (data) are inserted per user as memory items; by default, blobs are removed after processing unless persisted via configuration. A flush mechanism moves buffered data into memory, with sync options to wait for processing. Memory can be packed into prompts using a context API that returns a formatted string for prompting. The system uses a buffer to limit hot-path memory, and triggers a flush when the buffer grows too large or idle for a period.
Getting started
- Start the Memobase server locally or use Memobase Cloud with a free tier.
- Ensure you have a project URL and a project token (local: http://localhost:8019, secret; cloud: https://api.memobase.dev, sk-proj-xxxxxx).
- Install the Python SDK:
pip install memobase. - Follow the Python-oriented quickstart and API/docs for other languages.
Example usage snippets are provided in the README showing client creation, user management, data insertion with ChatBlob, flushing, and context generation. The quickstart script is available in assets and cross-language guidance exists for OpenAI/Ollama MCP integrations.
Recent releases
Latest version notes are shown under "Recent Updates" in the repository; 0.0.40 and earlier entries update LLM call counts and search performance, with notes about token cost reductions and improved event gist search.
Traction
The repository contains 2825 stars and 226 forks as its raw metrics. It has 30 open issues and is primarily a Python project under the Apache-2.0 license. Created in 2024-09-03 with the last push on 2026-01-11.
Behind the repo
Memobase provides multi-language clients (Python, Node, Go) and is marketed with a production-ready stack (FastAPI, PostgreSQL, Redis) and a Dockerized deployment. Documentation references include quickstart, API prompts, and MCP integration, with a live playground and UI tools mentioned in the README.
Caveats
License: Apache-2.0. Open issues as listed: 30. Age: created 2024-09-03; last push 2026-01-11. The README notes default data removal after processing unless configured to persist, and mentions a buffer-induced flush mechanism, with optional sync behavior.