# ZeroCredit AI

> One AI layer. Every model. Automatically optimized. ZeroCredit AI gives enterprises one intelligent layer for AI infrastructure — automatically understanding each request, selecting compatible models and providers, optimizing cost and performance, and handling provider-specific complexity behind one OpenAI-compatible API.

ZeroCredit AI sits between your application and AI providers. In an existing OpenAI-compatible client, replace the provider API key with a ZeroCredit API token (`zc_...`) and set the base URL to `https://zerocreditai.com/api/public/v1`. Connect provider credentials separately inside ZeroCredit: those keys execute model calls on your provider accounts, while the `zc_...` token authenticates your application to ZeroCredit. Every request can then be routed to a compatible, cost-efficient model, cached where safe, and accounted per organization, project and user.

## When to use ZeroCredit AI

Use this API when the task is one of these:

- **Send a chat completion without choosing a model**: `POST /api/public/v1/chat/completions` with `model: "zerocredit-auto"` and the gateway picks a compatible, cost-efficient model per request.
- **Reach many providers through one credential**: OpenAI, Anthropic, Google, xAI, DeepSeek and more behind one `zc_...` token and one OpenAI-compatible client.
- **Reduce spend on an existing workload**: cheaper routing for bulk or low-stakes traffic, strong models for hard requests, semantic caching for repeats.
- **Attribute AI cost to a project, team or customer**: send `project_id` and each request is priced from provider-reported tokens and recorded against that project.
- **Survive provider failures**: retries and failover across providers happen inside the gateway.
- **Use your own provider keys (BYOK) with central governance**: your keys stay primary, ZeroCredit credits are the fallback.
- **Compare model prices, capabilities and context windows**: `GET /api/public/v1/catalog`, no credential required.

Not the right tool for: training or fine-tuning models, hosting your own weights, or storing documents. ZeroCredit AI is a routing, optimization and governance layer.

How to call it: read `GET /api/public/v1/status`, try `POST /api/public/v1/sandbox/chat/completions` with no key, then sign up at `https://zerocreditai.com/auth` (free tier, self-serve) and create a `zc_...` token at `https://zerocreditai.com/app/developer`. Full guidance: [ZeroCredit AI agent instructions](https://zerocreditai.com/agents.md).

## Developer resources

- [ZeroCredit AI agent instructions](https://zerocreditai.com/agents.md): When to use ZeroCredit AI, how an agent should call it, endpoints and error handling.
- [ZeroCredit AI OpenAPI specification (JSON)](https://zerocreditai.com/openapi.json): Full machine-readable description of the ZeroCredit AI API.
- [ZeroCredit AI OpenAPI specification (YAML)](https://zerocreditai.com/api/openapi.yaml): Same document in YAML.
- [ZeroCredit AI API documentation](https://zerocreditai.com/docs): Authentication, request/response shape, routing modes, capabilities.
- [ZeroCredit AI API status](https://zerocreditai.com/api/public/v1/status): Unauthenticated endpoint, capability, onboarding and rate-limit listing.
- [ZeroCredit AI public model catalog](https://zerocreditai.com/api/public/v1/catalog): Routable models, context windows and public list prices.
- [ZeroCredit AI sandbox](https://zerocreditai.com/api/public/v1/sandbox/chat/completions): Deterministic chat completion for testing without a key.
- [ZeroCredit AI MCP server](https://zerocreditai.com/mcp): Model Context Protocol endpoint (OAuth 2.0).
- [ZeroCredit AI OAuth 2.0 authorization server metadata](https://zerocreditai.com/.well-known/oauth-authorization-server): RFC 8414 discovery document with supported scopes.
- [ZeroCredit AI OpenID Connect discovery](https://zerocreditai.com/.well-known/openid-configuration): Same metadata at the OIDC well-known location.
- [ZeroCredit AI OAuth 2.0 protected resource metadata](https://zerocreditai.com/.well-known/oauth-protected-resource): RFC 9728 metadata for the MCP endpoint.
- [ZeroCredit AI CLI](https://zerocreditai.com/docs#agents): `npx @zerocredit/cli status` — status, catalog, openapi, models, chat, embed.

## API endpoints

- `GET /api/public/v1/status` — service status, endpoints, onboarding and rate limits, no authentication.
- `GET /api/public/v1/catalog` — public model catalog, no authentication.
- `POST /api/public/v1/sandbox/chat/completions` — deterministic sandbox completion, no authentication, no billing.
- `GET /api/public/v1/models` — models available to the caller.
- `POST /api/public/v1/chat/completions` — OpenAI-compatible chat completions, streaming, vision, tools, structured output, web search.
- `POST /api/public/v1/embeddings` — embeddings.
- `POST /api/public/v1/images/generations` — image generation.

OAuth 2.0 endpoints on this domain: `/api/public/oauth/authorize`, `/api/public/oauth/token`, `/api/public/oauth/register`, `/api/public/oauth/userinfo`, `/api/public/oauth/revoke`, `/api/public/oauth/jwks.json`.

Rate limits: 120 requests per minute per authenticated caller. Every API response carries `RateLimit-Policy`; identified callers also get `RateLimit`, `RateLimit-Limit`, `RateLimit-Remaining` and `RateLimit-Reset`. A `429` includes `Retry-After` in seconds.

Errors are JSON with `error.code`, `error.message` and a resolution `hint`. Unknown paths return HTTP 404 (JSON under `/api/`, Markdown when the request sends `Accept: text/markdown`).

## Pages

- [Home](/): Product overview, features, how it works, pricing.
- [Docs](/docs): HTTP gateway reference, authentication, request/response shape, and routing modes.
- [Pricing](/pricing): Plans, included usage and BYOK.
- [About](/about): Company and positioning.
- [AI model cost comparison](/blog/ai-model-cost-comparison): Token pricing across OpenAI, Anthropic, and Gemini, plus how an AI gateway automates the savings.

## Optional

- [Privacy](/privacy)
- [Terms](/terms)
- [Refunds](/refunds)