Thanks for your interest. MemOS is an early-stage project and contributions are welcome.
git clone https://github.com/ajithhraj/MemOS.git
cd MemOS
python -m venv .venv
# Windows
.venv\Scripts\Activate.ps1
# Mac / Linux
source .venv/bin/activate
pip install -e ".[dev,llm,vector]"
cp .env.example .envSet ANTHROPIC_API_KEY in .env if you want to run the full extraction pipeline. Without it, MemOS falls back to the lightweight keyword extractor.
pytestCore tests in tests/ cover the decay math, extraction fallback, and end-to-end retrieval. They run without an API key or a running server.
- Memory consolidation — merge near-duplicate nodes (cosine similarity > 0.92) into a single enriched node.
- Conflict detection — flag when a new memory contradicts an existing one, surface it to the user.
- Export formats — add a JSONL export alongside the existing JSON export for streaming-friendly use cases.
- Dashboard polish — the React dashboard is functional but sparse. Better node tooltips, a timeline view, and a stats panel are all welcome.
- More tests — especially for the FastAPI routes and the SSE events stream.
- Keep PRs focused on one thing. A PR that fixes a bug and adds a feature is hard to review.
- Add or update tests for any behaviour change.
- Run
pytestbefore pushing and make sure it passes. - If you are adding a new dependency, update both
pyproject.tomlandrequirements.txtand explain why it is needed.
Open a GitHub issue with a description of the problem, the Python version, OS, and the full error traceback if applicable.