Introduction

Run open-weight models through OpenAI- and Anthropic-compatible endpoints.

Prism is an inference API for coding agents. Use the OpenAI client already in your application, change its base URL, and select a Prism model.

Zero data retention on every request. Prism does not store API inputs or outputs after processing or use them for model training. See exactly what ZDR covers.

OpenAI base URL: https://api.prisminference.com/v1

Prism currently serves DeepSeek-V4.1-Flash and DeepSeek-V4-Flash. GLM-5.3, Kimi K3, and Qwen are in progress. The same API key works across the supported wire formats.

API formats

Endpoint	Format	Purpose
POST /v1/chat/completions	OpenAI Chat Completions	Generate from text or images, stream tokens, call tools, and return structured output.
POST /v1/responses	OpenAI Responses	Use stateless Responses clients, tools, image input, and typed streams.
POST /v1/messages	Anthropic Messages	Use Anthropic clients, content blocks, tools, images, and streaming.
POST /v1/messages/count_tokens	Anthropic Messages	Count text and tool input tokens before inference.
GET /models	OpenAI Models	List available models.
GET /models/{model}	OpenAI Models	Retrieve one available model by ID.
GET /health	JSON	Check API availability.
GET /	JSON	Read API metadata and discovery links.

DeepSeek-V4.1-Flash accepts text and image input. DeepSeek-V4-Flash accepts text only. See Models for the current capability matrix.

OpenAI-compatible request

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-flash",
  messages: [
    { role: "user", content: "Find the race condition in this worker: ..." },
  ],
});

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

Keep API keys on your server. Do not put them in browser code, client bundles, or public repositories.

Start building

Create an API key

Create an account and issue a key from account settings.

Quickstart

Send a request with cURL, TypeScript, or Python.

Models

Compare model IDs, context windows, and intended workloads.

Chat Completions

Read the OpenAI-compatible request and response contract.

Streaming

Process incremental Chat Completions output.