Three steps. Thirty seconds.
No servers to run, no config files to edit, no long manuals.
Sign up and get your key
Create an account. The key shows up right away and unlocks the model groups on your plan.
~10 secondsSet one environment variable
Configure your tool to use our endpoint. This is typically a single line in your shell profile. Most Anthropic-compatible clients will use it automatically.
~10 secondsCode like normal
Open the tool you use—OpenClaw, Hermes, Pi, Conductor, OpenCode, Claude Code CLI, Codex Desktop, or Cursor—and it routes requests through the key.
~10 secondsOne line in your shell profile.
Add this to your shell profile (~/.zshrc, ~/.bashrc, etc.):
export ANTHROPIC_BASE_URL=https://openlimits.app
export ANTHROPIC_API_KEY=your-key-hereAfter setting the variables, continue using Claude Code CLI, Conductor, Cursor, or any other Anthropic-compatible client as usual. For complete API documentation, see the API Docs.
Setup by tool
The shell variables cover most tools. Here are the notes for specific ones.
Claude Code CLI
Add the two lines to your shell profile and restart the terminal. Every session routes through the key. Extended thinking and streaming keep working.
Conductor
It reads the Claude Code CLI settings when they exist. Otherwise set ANTHROPIC_BASE_URL and ANTHROPIC_API_KEY in Settings → Env.
Cursor
Add OpenLimits in the API provider section. Set the base URL and paste the key. The models on your plan appear in the list.
Python SDK
Pass base_url="https://openlimits.app" when you create the client, or set the env vars. The SDK reads them. OpenAI SDK works the same way on /v1/chat/completions.
TypeScript SDK
Pass baseURL: "https://openlimits.app" to the constructor. Env vars work the same. Streaming and extended thinking are supported.
cURL / Direct API
Swap the host for openlimits.app. Send the key with x-api-key or the Bearer header.
What happens when you send a request
OpenLimits functions as a lightweight proxy. Requests are routed to an appropriate account for the requested model. The additional processing typically completes within the first 50 ms.
Edge receives your request
The request reaches the OpenLimits application service. We authenticate the key and validate the request format.
Provider selected
The request is assigned to one of the eligible provider accounts for the model. If that account is unavailable or rate-limited, the system automatically retries with another eligible account.
Proxied and streamed back
The request is forwarded to the actual provider using its credentials. The response streams back in real time. On certain errors, the system switches accounts and retries automatically.
Works with your existing setup
It is a drop-in for the Anthropic endpoint. No code changes, no new packages, no prompt edits. Your tools and scripts keep the same shape. The difference is the rate limits.
Model groups behind one API key
GLM, MiniMax, DeepSeek, Claude, Codex, GPT-5.6, and Fable come through one endpoint. Request and response shapes stay the same, so you do not need separate integrations for different agent workloads.
No vendor lock-in
Drop the environment variable and you are immediately back on the direct Anthropic API. No custom SDK or proprietary format.
Native model-group endpoint
/v1/models lists the available groups. /v1/chat/completions keeps OpenAI-compatible clients working on the same host.