# AI Setu

> AI Setu is an OpenAI-compatible AI gateway: one API key routes chat, streaming, and embeddings calls to multiple LLM providers, with model-based routing, bring-your-own-key (BYOK), and cross-provider failover before the first byte.

## API

- Base URL: `https://gateway.aisetu.ai/v1`
- Auth: `Authorization: Bearer <workspace-api-key>`
- OpenAI-compatible: point any OpenAI SDK at the base URL above (change only `baseURL` and the key) and it works unchanged.
- Endpoints:
  - `POST https://gateway.aisetu.ai/v1/chat/completions` — OpenAI-canonical chat completions, including `stream: true` SSE.
  - `POST https://gateway.aisetu.ai/v1/embeddings` — OpenAI-canonical embeddings.
  - `POST https://gateway.aisetu.ai/v1/messages` — Anthropic Messages API-compatible chat. Rejects Bedrock/Gemini/Vertex models (400) — use `/chat/completions` for those.
  - `POST https://gateway.aisetu.ai/v1/prompts/{ref}/completions` — Render a managed prompt (by id, slug, or `slug@version`/`slug@label`) and run it as a chat completion.
- Providers: OpenAI, Anthropic, Google Gemini, AWS Bedrock, Azure OpenAI, Vertex AI have dedicated gateway adapters. Groq, xAI, DeepSeek, Mistral, vLLM & Ollama are reachable today via a BYOK connection using the OpenAI-compatible base-URL override (no bespoke adapter needed for those).

## Docs

- [Quickstart](https://docs.aisetu.ai/getting-started/quickstart) — Get credentials, first inference call
- [Providers & routing](https://docs.aisetu.ai/concepts/routing)
- [BYOK](https://docs.aisetu.ai/concepts/byok)
- [Budgets & spend limits](https://docs.aisetu.ai/concepts/budgets)
- [Guardrails](https://docs.aisetu.ai/concepts/guardrails)
- [Prompt management](https://docs.aisetu.ai/concepts/prompts)
- [Load balancing](https://docs.aisetu.ai/concepts/load-balancing)
- [Ephemeral tokens](https://docs.aisetu.ai/concepts/ephemeral-tokens)
- [Multi-tenant provisioning](https://docs.aisetu.ai/concepts/provisioning) — One-call child-tenant creation (control plane, not the gateway)
- [Errors](https://docs.aisetu.ai/operations/errors)
- [Rate limits](https://docs.aisetu.ai/operations/rate-limits)
- [MCP server](https://docs.aisetu.ai/integrations/mcp)
- [SDKs & CLI](https://docs.aisetu.ai/integrations/sdks)

## Full reference

- [llms-full.txt](/llms-full.txt) — request/response examples, routing & failover, BYOK, caching, budgets, guardrails, prompt management, load balancing, ephemeral tokens, error codes