# Prism Inference

> Prism serves open-weight models through OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages APIs.

## API

- OpenAI base URL: https://api.prisminference.com/v1
- API key environment variable: `PRISM_API_KEY`
- `POST /v1/chat/completions`: OpenAI Chat Completions
- `POST /v1/responses`: OpenAI Responses
- `POST /v1/messages`: Anthropic Messages
- `GET /v1/models`: list the public model catalog
- `GET /v1/models/{model}`: retrieve one model

Authenticate with `Authorization: Bearer PRISM_API_KEY`. Keep the key on the
server.

## Data handling

- Strict zero data retention for API inputs and outputs.
- API inputs and outputs are never used for model training.
- [Zero data retention](https://docs.prisminference.com/zero-data-retention)
- [Privacy policy](https://prisminference.com/privacy)
- [Terms of service](https://prisminference.com/terms)

## Models

| Model | Model ID | Status | Context | Intended workload |
| --- | --- | --- | --- | --- |
| DeepSeek-V4.1-Flash | `deepseek-v4.1-flash` | Available | 1M | Text and image input for coding, vision, reasoning, and tool use |
| Gemma 4 31B | `gemma-4-31b` | Available | 32K | Text and image input at BF16 for vision, coding, and tool use |

Use `GET /v1/models` to read the callable public catalog. No API key is
required. Do not send requests for models marked in progress.

## OpenAI example

```typescript
import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.PRISM_API_KEY,
  baseURL: "https://api.prisminference.com/v1",
});

const response = await client.chat.completions.create({
  model: "deepseek-v4.1-flash",
  messages: [
    { role: "user", content: "Review this TypeScript diff: ..." },
  ],
});

console.log(response.choices[0].message.content);
```

## Agent integration

Prism uses prompt-first setup and provides an immediate-key signup API for
agents.

1. Read https://docs.prisminference.com/guides/coding-agents and choose the exact recipe for the
   current coding agent.
2. Inspect existing provider configuration without reading or printing secrets.
3. If no key is available, collect the owner email and username, then send
   `human_email` and `username` to `POST /api/agent-signups`.
4. Store the returned one-time `api_key` immediately through the harness's
   native secret mechanism. Never repeat it to the human or include it in the
   final response.
5. Ask the human only for the emailed OTP. Verify through `POST
   /api/agent-signups/verify` with the key as Bearer authentication.
6. Show a redacted preview and make only the targeted provider change.
7. Query `GET /v1/models` and validate the configured provider and model.

## Documentation

- [Introduction](https://docs.prisminference.com/introduction): Supported formats, base URLs, and the 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, and key handling.
- [SDKs and clients](https://docs.prisminference.com/sdks): OpenAI, Anthropic, and agent-harness clients.
- [Rate limits](https://docs.prisminference.com/rate-limits): Per-key limits and `Retry-After`.
- [Errors](https://docs.prisminference.com/errors): Error envelopes and retry policy.
- [Chat Completions](https://docs.prisminference.com/api-reference/chat-completions)
- [OpenAI Responses](https://docs.prisminference.com/api-reference/responses)
- [Anthropic Messages](https://docs.prisminference.com/api-reference/messages)
- [OpenAPI 3.1](https://docs.prisminference.com/openapi.yaml)
- [Streaming](https://docs.prisminference.com/guides/streaming)
- [Tool calling](https://docs.prisminference.com/guides/tool-calling)
- [Structured outputs](https://docs.prisminference.com/guides/structured-outputs)
- [Coding-agent setup](https://docs.prisminference.com/guides/coding-agents)
- [Codex](https://docs.prisminference.com/guides/codex)
- [Cursor](https://docs.prisminference.com/guides/cursor)
- [OpenCode](https://docs.prisminference.com/guides/opencode)
- [Hermes](https://docs.prisminference.com/guides/hermes)
- [OpenClaw](https://docs.prisminference.com/guides/openclaw)
- [Documentation index](https://docs.prisminference.com/llms.txt)
- [Full machine context](https://prisminference.com/llms-full.txt)
- [Agent Skill](https://prisminference.com/skills/prism-inference/SKILL.md)

## Public pages

- [Home](https://prisminference.com)
- [Models](https://prisminference.com/models)
- [Pricing](https://prisminference.com/pricing)
- [Pricing (machine-readable)](https://prisminference.com/pricing.json)
- [Status](https://status.prisminference.com)
- [Docs](https://docs.prisminference.com)
- [Create an account and API key](https://prisminference.com/signup)

## Machine-readable

- [OpenAPI 3.1](https://docs.prisminference.com/openapi.yaml)
- [API catalog](https://prisminference.com/.well-known/api-catalog)
- [Security contact](https://prisminference.com/.well-known/security.txt)

Company: Prism Technologies Inc. Backed by Y Combinator.