# Tether

> Tether is the governed execution layer between AI and your application: a headless
> HTTP + SSE service that routes intent, calls a model with the customer's registered
> tools over a signed webhook, stages every write for explicit human confirmation, and
> returns typed UI blocks plus per-turn cost accounting. It never receives database
> credentials, and it never executes a write directly.

Key facts:
- Headless: no UI ships with it; clients render typed blocks.
- Tools are registered as JSON Schema; whether a tool writes is declared, never inferred.
- Writes are staged as pending actions; a separate authenticated confirmation executes
  them exactly once (compare-and-set claim; never retried).
- Tenant scope comes from the API key, never from request bodies or model arguments.
- Spend: append-only ledger, per-tenant budgets, fail-closed pricing, pre-call breaker.
- Single model provider today. Pre-launch; design partners onboarded manually.

## Docs
- [Documentation](/docs): full documentation index
- [Introduction](/docs/introduction): what Tether is, is not, and when not to use it
- [Quickstart](/docs/quickstart): catalog → endpoint → first answer
- [Your first write](/docs/first-write): the staging and confirmation flow end to end