projects /
Poundlock Toolkit
Find out which local models, runtimes, and settings actually work on your hardware, then compare the results.
Mockup: copy condensed from the project README. Review before publishing.
Run Toolkit on a Linux or macOS machine, or point it at one over SSH. It inspects the hardware, runs the tests you approve, and keeps the results so you can compare setups later. It doesn’t need a server or a model gateway, and it doesn’t install models or drivers: you choose what to test.
poundlock-toolkit inspect
poundlock-toolkit plan examples/openai.json --output plan.json
poundlock-toolkit recommend demo.json --output report.md- Plans before execution. Review the bounded test plan before anything runs with
--execute, and keep the receipt. - Tool-call schema checks. Tests nested tool-call JSON through native Ollama and OpenAI-compatible endpoints and reports which field failed, without saving the response text.
- Honest results. Synthetic tests never stand in for real hardware or model quality.
A write-up of the research behind Toolkit will appear in the journal.