As AI becomes more prevalent in software, testing becomes more important — not less. Dogfood Lab is an open workshop for the testing problem: new methodologies, operating systems for running tests, and public showcases of results.
Sister org to MCP Tool Shop. Where MCP Tool Shop builds local-first AI tooling, Dogfood Lab stress-tests how AI-assisted software should be verified.
Intentionally open-ended. Early directions:
- Test operating systems — frameworks that coordinate test runs across agents, tools, and humans.
- New methods of testing — property-based proofs for LLM pipelines, visual regression for diffusion outputs, dogfood-driven quality gates, adversarial audits.
- Showcases — public writeups of what worked, what broke, what got shipped.
If an idea needs a home and it's testing-shaped, it lives here.
- Shipped — testing tools and operating systems with CI, tests, releases, and install commands.
- Active Lab — in-progress experiments with daily commits.
- Seed Vault — archived prototypes and revival candidates.
| Dogfood-first | Every tool gets tested against itself before release. |
| Evidence over estimates | Showcase artifacts, not self-reports. |
| Local-first | Runs on your hardware. No cloud required. |
| Open-ended | Testing is a big word. Scope won't be narrowed prematurely. |
Questions or ideas? Open a Discussion Bugs go in Issues on the relevant repo.
Eat first. Ship second.