# gm — say gm to AI

> Privacy-first inference gateway on Bittensor. Point your existing OpenAI,
> Anthropic, or Gemini SDK at the gm gateway and get identical behavior — every
> request is sealed inside an Intel TDX TEE that keeps the route confidential
> in-transit and in-use before it reaches the model. Closed beta.

gm is OpenAI-, Anthropic-, and Gemini-compatible: keep your SDK, change the base
URL and the key, and your code works unchanged. Auth is a gm API key
(`Authorization: Bearer <gm-key>`), minted in the dashboard. Billing is prepaid —
no subscription. The enclave seals each request in-transit and in-use; gm does
not make retention or training claims about the upstream model makers.

## API

- Base URL: `https://api.saygm.com/v1`
- Auth: `Authorization: Bearer <gm-api-key>`
- OpenAPI spec: [openapi.json](https://saygm.com/openapi.json)
- `POST /v1/chat/completions` — OpenAI Chat Completions (streaming and non-streaming). Set your OpenAI SDK `base_url` to `https://api.saygm.com/v1`.
- `POST /v1/responses` — OpenAI Responses API.
- `POST /v1/messages` — Anthropic Messages API. Set the Anthropic SDK `base_url` to `https://api.saygm.com`.
- `POST /v1beta/models/{model}:generateContent` — Gemini. Set the Google GenAI client base URL to `https://api.saygm.com`.

## Models

- Send a bare model id (no provider prefix) as the `model` field; the gateway resolves the provider from its catalog.
- Frontier: proxied closed models from OpenAI, Anthropic, and Google (e.g. `gpt-5.5`, `claude-opus-4-8`, `gemini-2.5-pro`). The enclave seals the route, and the model maker receives the prompt under its own API terms.
- Confidential: open-weight models served inside a TEE (the `-TEE` ids), so inference itself runs sealed and the prompt stays inside the enclave end-to-end.
- Live pricing and the current model set are on the [pricing section](https://saygm.com/#pricing) of the home page.

## Guardrails

- Optional, per-API-key processors that run inside the enclave before the prompt leaves it.
- PII redaction, secret/credential redaction, custom regex, and blocked-word filters.
- Configured per key in the dashboard; off by default, opt-in per key.

## Billing

- Prepaid credit balance — no subscription, no free tier.
- Top up by card (Stripe), USDC, or x402 (alpha).
- Usage is metered per token; prices are at or below what the model maker charges.

## Links

- [Dashboard / sign-in](https://saygm.com)
- [Partners](https://saygm.com/partners)
- [OpenAPI spec](https://saygm.com/openapi.json)