data & privacy

routing decision only — control plane. The trust contract — verify it programmatically at /transparency?format=json.

What the edge sees

only the first ~2000 chars of your prompt (≈512 tokens, the bge-base-en routing model's limit) — used ONLY to pick a model. This is a routing PEEK, not a cap on your prompt: your FULL prompt runs on YOUR infra / frontier, never limited or stored by us.

What we log

per-license daily request COUNT (use:<key>:<YYYYMMDD>)
license metadata (plan/limit/email)

What we NEVER log

prompt content
your data
your API keys
model outputs

Where your data lives

your API keys, inference, and data stay on YOUR infra. We return only {route, probability, margin}.

Using your data to improve

Default: PRIVATE for everyone (all plans). Your prompt routes a request in-the-moment; it is NOT stored or used to train any central/shared model.

Self-improvement: happens LOCALLY on your own instance (your traffic improves your router); your data never leaves your infra.

Opt-in: you may OPT IN to contribute anonymized routing labels (the route decision, not raw prompts) to improve the shared router — for account credit. Never opt-out, never default.

edge model: @cf/baai/bge-base-en-v1.5 (embedding only, at the edge) · open edge worker + SDKs, auditable; logs zero prompt content (most requests never leave your infra)