Independent AI lab · Public instrument
Explore Failure - Harness Workbench
Give your local model and coding agent a real job. Harness Workbench checks the result independently, keeps the evidence, and saves a Recipe you can run again. The agent saying “done” doesn’t earn a pass.
View the repository · Read the documentation
Verified release facts
| Primary purpose | Try local AI on real jobs and verify what actually happened |
|---|---|
| Measurement method | Independent checks on task outcome, file changes, and agent evidence; controlled experiments that change conditions and introduce faults |
| Supported systems | Linux, macOS; POSIX-oriented; Windows unsupported |
| Python support | CPython 3.11–3.14 |
| Runtime dependencies | None |
| AI model required | For the Reality Check (local Ollama); not for the experiment engine |
| Interface | Command-line application |
| Evidence produced | Verdict, verification facts, changed files, sealed run record, and a rerunnable Recipe |
| License | Apache-2.0 open source |
| Current release | v0.2.0 on GitHub; not published to PyPI |
The four-stage measurement cycle
Harness Workbench gives the agent a real task, runs it on your hardware, checks the result independently, and saves a Recipe you can run again.