SmartRouter is a model-routing policy and local supervision toolkit for coding agents. It makes permissions, capabilities, verification, budget, and routing reasons inspectable. The live-provider tools remain experimental; use the README's local checks to begin without an API key.
- Read the project introduction.
- Ask integration questions in Discussions.
- Pick a LLM testing issue.
- Begin with unseen reservation fixtures if you want a task that needs no provider account or paid calls.
- Follow CONTRIBUTING.md for setup, verification, and pull requests.
When sharing results, include the model/profile identity, task packet, acceptance checks, failed runs, and available usage metadata. Label missing measurements as unknown. A synthetic policy test is not proof of production savings or reliability.
A useful introduction explains a concrete routing problem and links to the quick start. A walkthrough should show one passing case and one automatic refusal, including the evidence that supports the outcome. Contributions are especially useful when they add new cases or reproduce a bug.
Use the visual README preview alongside a short explanation of what the project currently does. Link a specific issue when asking for help so readers can see the scope and acceptance criteria. Disclose your connection to the project and AI assistance where required by the destination's rules.
The project already has a pending awesome-list submission. Acceptance belongs to that list's maintainers. Other directories have their own requirements; check them before submitting and do not claim a listing is accepted until it appears upstream.
If you find SmartRouter helpful, please consider giving it a star. Reproducible feedback and focused contributions are welcome too.