Skip to content

Prism documentation

Prism is a self-hosted AI gateway. You run it on your own machine or server, connect your model providers once (API keys, subscriptions or your own model servers), and point your applications at one address. Prism chooses a model for each request, fails over when a provider is down or out of quota, and records what each request cost.

Prism speaks the OpenAI, Anthropic Messages, Gemini and Ollama APIs, so most tools work without code changes: Claude Code, Codex CLI, Gemini CLI, the OpenAI and Anthropic SDKs, Open WebUI and other Ollama clients.

  1. Install Prism with one command on Linux or macOS.
  2. Quick start: add a provider in the web console and send a first request.
  3. Connect your clients: settings for each tool.
  • Web console: providers, routing, agents, usage and updates in the browser.
  • Managed inference: start, stop and power your own model servers from Prism (Pro and above).
  • Privacy and data: what Prism stores, what it sends and how to turn each part off.