What is OpenLLM?

OpenLLM is an AI gateway. It sits between your AI subscriptions and the tools you work in, and lets you use any model from any client through a single endpoint.

Most people already pay for one or more AI subscriptions: Claude, ChatGPT, Grok, Kimi. Each one normally locks you into its own app or CLI. OpenLLM connects those subscriptions (and plain API keys) to a single gateway, so the model you're entitled to and the tool you prefer no longer have to come from the same company.

Use any CLI or client you want

Point your favorite coding agent at OpenLLM and pick a model:

• Claude Code
• Codex CLI
• OpenCode
• Kimi Code
• Raycast
• Hermes

Want Claude Code running on GPT, or Codex running on Claude? That's the point. One gateway, your choice of model on either side.

One chat for all your subscriptions

The OpenLLM web app includes a unified chat window where you can talk to every model you're connected to from one place. Compare answers, switch models mid-conversation, and stop juggling four different chat tabs for four different subscriptions.

Smart routing and fallbacks

OpenLLM routes requests across your providers with tier aliases (for example ultra, plus, lite), so your tools ask for a capability level instead of a hardcoded model name. If a provider is down or you hit a quota, fallback chains keep your session going on the next available option. Provider quotas and terms change often; a gateway means no single provider can strand you.

Unified memory, context, and codebases, on every device

Because everything flows through one gateway, your context comes with you:

• Unified memory: what you've taught one assistant is known to all of them, instead of being trapped inside one vendor's app.
• Shared context: conversations and project context carry across clients and models.
• Indexed codebases: index a repo once and every connected client and model can search it.
• Multi-device: the same memory, context, and indexes follow you from your laptop to the web app to any other machine you sign in on.

Private by design

Credentials live in an encrypted vault with zero-knowledge encryption. Your recovery phrase is the only way in; OpenLLM cannot read your keys or recover the vault for you.

Not the model-serving library

Searching for the open-source model-serving project by BentoML? That's a different OpenLLM. This one, at openllm.sh, is a gateway for using your AI subscriptions anywhere.

Get started

Sign up at openllm.sh, connect a subscription or API key, and point your CLI at the gateway. Docs live at docs.openllm.sh.