Features
What you get for changing one line
Clawrouter sits between your code and every model provider. The interface stays the one you already use; what changes is the routing, the visibility and the control.
One endpoint. Every provider.
OpenAI-compatible by design. Keep your SDK, your code, your habits.
Point the base URL at Clawrouter, name a model, and the request goes out. No adapter layer to maintain, no per-provider branch in your code, and no rewrite when you want to try something new — a different model is a different string.
The best route wins. Every request.
Price, latency and health are watched across providers; each request takes the fastest cheap path.
Channels serving a model are tried from the highest priority tier down, weighted at random within a tier. A failing channel falls through to the next candidate rather than failing your request, and a channel that keeps failing is taken out of rotation automatically.
See every token.
Cost, latency and usage per model and per key, tracked live.
Every call is logged with its own cost breakdown: uncached input, cache reads, cache writes and output, each with the ratio applied and the dollars it came to. Open a row and the number is explained instead of asserted — no mystery bill at the end of the month.
Keys with rules.
Budgets, model allowlists and IP restrictions per key.
Hand your team keys, not your credit card. Each key carries a hard spending cap, an expiry, and the exact set of models it may call. Organizations go further: the org holds its own balance, keys bill against it, and an owner can allocate budget to a member and pull back what they have not spent.