Installation¶

Python 3.10 or newer. The database is embedded, so there is nothing to install or run beside the package itself.

From PyPI¶

uv add audit-ready-memory

Check it worked:

uv run python -c "from audit_ready_memory import Memory; print('ok')"

Versions are below 0.1.0 on purpose: the API is usable but not frozen. Pin an exact version if you depend on it.

Without installing anything¶

To try it in a throwaway environment that leaves nothing behind:

uv run --with audit-ready-memory python

From source¶

For the demo, the tests, or to work on the library itself:

git clone https://github.com/humemai/audit-ready-memory
cd audit-ready-memory
uv sync

Optional extras¶

The demo needs Streamlit and the file parsers:

uv sync --extra demo

Everything, including test and documentation tooling:

uv sync --all-groups --all-extras

Dependencies¶

Two at runtime:

arcadedb-embedded for the local database. It bundles a JVM, so the wheel is large, around 63 MB. There is no server component.
pydantic for the event schema.

The demo extra adds streamlit, openai, pandas, openpyxl, pypdf, and python-dotenv.

Running the tests¶

uv run pytest

Configuration¶

The library itself needs no configuration. The demo reads OPENROUTER_API_KEY from the environment or a .env file, and runs without it with the AI chat disabled. Copy .env.example to .env to set one. The key is never written to the memory database or the log.

Where data goes¶

Memory("./memory-db") creates and uses that directory. It is the whole database: back it up, move it, or delete it as a unit. Restrict its permissions to the service user, because content is stored unencrypted.

Building the docs¶

uv sync --group docs
uv run mkdocs serve