Instead of being a simple memory database, memabra acts as a meta-cognitive controller for agents: given a task, it quickly decides whether to answer directly, recall memory, load a skill, or invoke a tool — and continuously improves this judgment based on task outcomes.

Install

git clone https://github.com/TacitLab/memabra.git
cd memabra
python -m venv venv
source venv/bin/activate
pip install -e ".[dev]"

Quick start

1. See the available commands

memabra --help

2. Run a dry-run evaluation

A safe way to see the full workflow without actually promoting a new router version:

memabra run --dry-run --format text

3. Check system status

memabra status --format text

4. List saved router versions

memabra version list --format text

5. Roll back to a previous version

memabra version rollback <version-id> --format text

CLI subcommands

Command	Description
`memabra run`	Run the online learning workflow
`memabra status`	Show current system state
`memabra version list`	List all saved router versions
`memabra version rollback <id>`	Roll back to a specific version

Text output format

By default, memabra prints JSON. For operator-friendly summaries, add --format text:

Status — current version, trajectory/report counts, latest report timing and promotion outcome.
Version list — total count, current active version highlighted.
Workflow — grouped into Summary, Baseline, Challenger, Deltas, and Decision sections with normalized yes/no flags and fixed-precision metrics.

Running tests

pytest tests/ -q

Project status

See docs/PROGRESS.md for a detailed capability roadmap and docs/DEMO.md for walkthrough examples.

License

MIT