Anvil Integration
Connect Anvil to LLM Gateway, choose DevPass or pay-as-you-go billing, and configure its local coding engine.
Anvil is a desktop workspace for repository-based agent work. Its built-in LLM Gateway connector supports browser authorization and API keys.
Video walkthrough
Install Anvil
Download the desktop app from the official releases. On first launch, choose your role and open Primary agent. You can also configure providers later in Settings.
Connect LLM Gateway
Select LLMGateway, then choose the billing mode matching your account:
| Mode | Use it with |
|---|---|
| DevPass | A coding plan and canonical models supported by that plan |
| Pay as you go | A workspace billed against gateway credits |
Click Connect in browser. Review the account and organization on the authorization page, then approve the connection while Anvil stays open. Alternatively, paste an existing key into API key (alternative).
The selected mode determines which model catalogue Anvil loads. Keep DevPass models canonical; provider-pinned routing is not available on coding plans.
Check the coding engine
Anvil uses a local Codex coding engine to edit files and run commands. Models and billing still come through LLM Gateway; this connector does not require a separate ChatGPT account.
If Anvil reports that the engine is missing or unavailable, use its Install coding engine or Repair coding engine control, then check the status again.
Select a model and test
After connecting, choose a tool-capable model from the validated catalogue and click Test Connection. Check capabilities on the models page.
Start a small task in your repository, review the proposed changes, and run the relevant tests. Inspect usage in the gateway workspace you authorized.
Troubleshooting
Browser authorization times out: start the connection again and finish the approval while Anvil remains open.
No models appear: check the connected key and billing mode, then refresh the connector status.
The coding engine is unavailable: use the install or repair control before starting an agent task.
Authentication fails: verify that the key is active, or reconnect in the browser to authorize a new connection.