Ship more without spending more
Route each request to the best quality-per-token model, right inside Claude Code, Codex and Cursor.
You don’t send a freight truck to deliver a postcard.
Most coding prompts don’t need a frontier model. Some absolutely do. The router reads each request, sends it to the smallest model that finishes the job at full quality, and saves frontier for the work that earns it.
Cost per merged PR against frontier-only routing.
Average code-quality score held across routed traffic.
Requests completed, with automatic failover.
Quality vs. Cost per Task
weave · routed
94
$
deepseek-v4
94
$
claude-opus-4.8
94
$$$$
claude-sonnet-5
88
$$$
gpt-5-codex
86
$$$
Quality-first
Frontier when it counts
Routes to frontier whenever scores separate. Set a floor the router never crosses.
Balanced · default
The parity-band rule
Cheapest model inside the band, frontier outside it. Where most teams stay.
Cost-first
Every token audited
Lowest-cost model that clears your bar, every request. For high-volume routine work.
one policy
routing pool
Get Started
Your questions and answers — start routing with weave
Frequently asked questions
Classification runs in under 50ms on a small distilled model that lives in our edge layer. The first token from the picked model still streams in the same envelope it would have without Weave, usually faster, because we route past the most-congested providers when we can.
Get your router key
Generate a bearer token. The token never leaves your device unless you choose to export it.
wv_•••• •••• a91f
Run one command
The installer detects your clients and points them at the router. One env var per provider, written automatically. Anthropic, OpenAI, and Google.
$ npx -y @workweave/router
Work as usual
Each request is now routed to the model with the best quality-per-token. The client is unaware of the proxy.
✓ routed via weave
Source code available. Start taking your token budget twice as far.
Get started in 5 minutes or book a demo with our team.
