Local-first alpha for agent builders

Agent Black Box records AI agent runs so you can debug them locally.

Capture a run, inspect the Debug Path, export support or handoff packets, and delete local trace data when you are done. No account or API key required for the demo path.

Download alpha zip View release Try a real workflow Read transcript

Checksum for agent-black-box-0.1.0-design-partner.zip:

20e3f38638c03fd24a94199ef2d058313214055c1a05a7fb67f678f5cc91db62

Terminal transcript showing Agent Black Box doctor, record, show, and delete commands

After download

Unzip the design-partner kit, install into its local virtualenv, and run the doctor before recording a demo run.

unzip agent-black-box-0.1.0-design-partner.zip
cd agent-black-box-0.1.0-design-partner
sh install.sh
. .venv/bin/activate
abb doctor

What the alpha proves

Record locally

Wrap a command or use the SDK and keep trace data on your machine by default.

Inspect the first clue

abb show summarizes runs, warnings, artifacts, and the first useful Debug Path.

Hand off safely

Create support and handoff packets before sharing full local bundles.

Clean up data

abb delete RUN_ID --yes removes local trace rows and unreferenced artifact objects.

Try it in 60 seconds

No account, cloud project, hosted backend, or package registry is needed for the first proof.

doctor checks local setup.
record captures the demo agent run.
show prints the run summary and Debug Path.
delete clears the local run data.

python3 abb.py doctor
python3 abb.py record --name demo -- python3 examples/basic_agent.py
python3 abb.py runs
python3 abb.py show RUN_ID
python3 abb.py delete RUN_ID --yes

Looking for three design partners

Best fit: agent founders, agent infra engineers, and local automation builders with real debugging pain who can spend 20 to 40 minutes on the alpha.