Contributing to archon-memory-core

Development install

git clone https://github.com/Divergence-Router/archon-memory-core.git
cd archon-memory-core
pip install -e ".[reranker,graph,dev]"

Running the tests

pytest tests/ -v

All tests are offline — no API keys or running services required.

Running the benchmark

The Agentic Memory Benchmark (AMB) covers 10 scenarios and 200 queries.

# Run against archon-memory-core (default adapter)
python benchmark/run_benchmark.py

# Run a single scenario
python benchmark/run_benchmark.py --scenario 01_personal_assistant

# Run against a different adapter
python benchmark/run_benchmark.py --adapter benchmark.adapters.naive_vector.NaiveVectorAdapter

Results are written to benchmark/results/.

Adding a new adapter

1. Create a new file in benchmark/adapters/, e.g. my_adapter.py.
2. Implement the three required methods:

class MyAdapter:
    def ingest(self, session: list[dict]) -> None:
        """Store all turns from a session."""
        ...

    def query(self, q: str, n: int = 5) -> list[str]:
        """Return up to n text chunks relevant to the query."""
        ...

    def reset(self) -> None:
        """Clear all stored state between scenarios."""
        ...

3. Pass it to the benchmark runner with --adapter my_adapter.MyAdapter.
4. Open a PR — include benchmark results in the description.

Code style

• No formatter enforced yet; match the style of the file you're editing.
• Type hints preferred. Docstrings on public methods.
• New features need at least one test in tests/test_basic.py.