# OpenClaw

> Configure the OpenClaw agent framework to use gateway models.

_Source: https://aiprimetech.io/docs/guides/openclaw/ · Home > Docs > Integration guides_

OpenClaw accepts Anthropic-compatible endpoints, so it needs the base URL and key and nothing else.

```bash
export ANTHROPIC_BASE_URL="https://aiprimetech.io"
export ANTHROPIC_API_KEY="sk-your-key"
```

If your build uses a config file rather than the environment, set the provider to Anthropic, the base URL to `https://aiprimetech.io` (no `/v1`), and the model to an id from [Models](/docs/getting-started/models/).


## Agent economics

Autonomous frameworks can run for a long time without supervision. Two safeguards are worth setting before the first long run:

- A hard step limit, so a stuck loop cannot spin indefinitely.
- A cheaper model for routine sub-tasks, reserving the frontier model for planning.

- [Cost control](https://aiprimetech.io/docs/guides/cost-control/) — Bounding an autonomous run
- [Rate limits](https://aiprimetech.io/docs/api-reference/rate-limits/) — Concurrency ceilings


---
_ClaudeAPIKey.dev is an independently operated, Anthropic-compatible API gateway. Not affiliated with Anthropic._