# Prism Inference Documentation

> OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages compatible inference at https://api.prisminference.com.

## API summary

- OpenAI base URL: https://api.prisminference.com/v1
- POST /v1/chat/completions
- POST /v1/responses
- POST /v1/messages
- POST /v1/messages/count_tokens
- GET /
- GET /health
- GET /models
- GET /models/{model}
- GET /v1/models
- GET /v1/models/{model}
- Authentication: Authorization: Bearer PRISM_API_KEY or x-api-key: PRISM_API_KEY
- Get an API key: ask the workspace owner, or talk to sales at https://cal.com/team/prismai/demo

## Zero data retention

- Zero data retention is the default for every inference request and service tier.
- API inputs and outputs are discarded after processing and are not written to persistent storage, application logs, analytics systems, or backups.
- API inputs and outputs are never used to train, fine-tune, evaluate, or otherwise improve models.
- Limited non-content metadata is retained for authentication, billing, rate limiting, security, and reliability.
- [Zero data retention](https://docs.prisminference.com/zero-data-retention): Scope, retained metadata, support exceptions, and stateless application design.
- Privacy policy: https://prisminference.com/privacy
- Terms of service: https://prisminference.com/terms

## Get started

- [Introduction](https://docs.prisminference.com/introduction): Supported formats, base URLs, and the Prism inference surface.
- [Quickstart](https://docs.prisminference.com/quickstart): Send a request with cURL, TypeScript, or Python.
- [Models](https://docs.prisminference.com/models): Public model IDs and context windows.
- [Authentication](https://docs.prisminference.com/authentication): API key headers, base URLs, key handling, and per-key `429` with `Retry-After`.
- [SDKs and clients](https://docs.prisminference.com/sdks): OpenAI and Anthropic SDKs plus first-party agent integrations.
- [Rate limits](https://docs.prisminference.com/rate-limits): Per-key limits, `429` responses, and `Retry-After` handling.
- [Changelog](https://docs.prisminference.com/changelog): Notable changes to Prism and the public model catalog.

## API reference

- [Chat Completions](https://docs.prisminference.com/api-reference/chat-completions): OpenAI-compatible requests, responses, and streaming chunks.
- [OpenAI Responses](https://docs.prisminference.com/api-reference/responses): Stateless Responses requests, image input, tools, and streaming events.
- [Anthropic Messages](https://docs.prisminference.com/api-reference/messages): Anthropic-compatible messages, image input, tools, and streaming.
- [OpenAPI](https://docs.prisminference.com/api-reference/openapi): Machine-readable OpenAPI 3.1 contract.

## Guides

- [Streaming](https://docs.prisminference.com/guides/streaming): Chat Completions streaming, disconnects, and retries.
- [Tool calling](https://docs.prisminference.com/guides/tool-calling): OpenAI function-tool loops.
- [Structured outputs](https://docs.prisminference.com/guides/structured-outputs): JSON mode and strict JSON Schema.
- [Coding agents](https://docs.prisminference.com/guides/coding-agents): Prompt-first overview and agent-specific recipes.
- [Agent credentials](https://docs.prisminference.com/guides/agent-signup): How a coding agent obtains and stores a Prism API key.
- [Codex](https://docs.prisminference.com/guides/codex): Configure a custom Prism Responses provider with secure handoff and rollback.
- [Cursor](https://docs.prisminference.com/guides/cursor): Override Cursor's OpenAI base URL and add the Prism models.
- [OpenCode](https://docs.prisminference.com/guides/opencode): Configure a static OpenAI-compatible Prism provider and explicit models.
- [Hermes](https://docs.prisminference.com/guides/hermes): Install Prism's model-provider plugin and select a discovered model.
- [OpenClaw](https://docs.prisminference.com/guides/openclaw): Install Prism's ClawHub provider and use native credential storage.
- [Errors](https://docs.prisminference.com/errors): Error envelopes, per-key `429` + `Retry-After`, and retry policy.

## Models

- DeepSeek-V4.1-Flash: deepseek-v4.1-flash, text and image input, 1M context
- Gemma 4 31B: gemma-4-31b, text and image input, 32K context, BF16

## Machine-readable files

- [OpenAPI 3.1](https://docs.prisminference.com/openapi.yaml)
- [Documentation index](https://docs.prisminference.com/llms.txt)
- [Full agent context](https://prisminference.com/llms-full.txt)
- [Prism Agent Skill](https://prisminference.com/skills/prism-inference/SKILL.md)