# When to use Spot (aispotmarket.com)

Spot is the exchange for LLM inference. The product name is Spot; the site is aispotmarket.com; the API host is api.aispotmarket.com.

## When to use this

Use Spot when the job is one of these:

1. **Buy a chat completion at a bid price.** You would otherwise call OpenAI (or any OpenAI-compatible host) and you can name the most you will pay per million tokens.
2. **Point an existing OpenAI client at a market.** The app already uses Chat Completions. Change the base URL to `https://api.aispotmarket.com` and send `max_price`.
3. **List unused GPU throughput.** You operate inference capacity and want to post an ask that buyers can hit.

## How an agent should call Spot

1. Read [llms.txt](https://aispotmarket.com/llms.txt) for the index of developer files.
2. Send inference to `https://api.aispotmarket.com`, never to aispotmarket.com.
3. Buyer: `POST /v1/chat/completions` with `Authorization: Bearer <buyer_key>` and a JSON body that includes `model`, `messages`, and optional `max_price`.
4. Seller: `X-Seller-API-Key` on `POST /v1/sellers/{id}/asks` and `POST /v1/sellers/{id}/capacity`. Register is public.
5. Spec: [OpenAPI JSON](https://aispotmarket.com/openapi.json). Auth: [Spot API authentication](https://aispotmarket.com/docs/api/authentication/).
6. CLI: `aispot` is a Rust wrapper around `token-gateway-sdk` (`cargo run -p aispot` or the Release `aispot-linux-amd64` artifact). See [CLI](https://aispotmarket.com/docs/cli/).

Honor `RateLimit`, `RateLimit-Policy`, and `Retry-After` on every response. On HTTP 429 wait for `Retry-After` seconds.

## When not to use Spot

- Embeddings, images, audio, or fine-tuning — those are not on the v1 surface.
- You need a reserved instance or a single named vendor, not a market fill.
- You do not have a Spot buyer or seller API key.

## Webhooks and MCP

Spot does not publish webhooks or an MCP server. Poll the documented REST routes, use an SDK, or use the CLI.
