Use your Claude Pro or Max plan as the model for an agent.
Overview
Pick one of the Claude Code models for an agent and its model calls run Claude Code on inference.sh, signed in with your own Claude plan.
| Model | App |
|---|---|
| Opus 5.5 | anthropic/claude-code-opus-5-5 |
| Sonnet 5.5 | anthropic/claude-code-sonnet-5-5 |
| Haiku 5.5 | anthropic/claude-code-haiku-5-5 |
| Fable 5.1 | anthropic/claude-code-fable-5-1 |
Claude Code only answers as the model. The agent's system prompt, tools and conversation history are inference.sh's. Tool calls Claude makes go back to your agent, which runs them the same way it does with any other model.
Prerequisites
- A Claude Pro or Max plan
- Claude Code installed on your computer
Step 1: Get a token
- Run
claudeand log in with your Claude account. - Run:
1claude setup-token- Copy the token it prints. It starts with
sk-ant-oat01-and is valid for one year.
Step 2: Connect in inference shell
- Go to vaults → credentials
- Find Claude Code (under Anthropic) and click Configure
- Paste the token
- Click Save credentials
The token is yours alone: it is saved as a personal credential, so other members of the workspace connect their own.
If you pick a Claude Code model before connecting, the agent chat shows a setup card with these steps and a link to the Claude Code credential page.
Step 3: Use it in an agent
Open the agent and choose one of the anthropic/claude-code-* models in its model picker. Runs of that agent now use your Claude plan.
Differences from the Anthropic API models
| Claude Code models | Anthropic API models | |
|---|---|---|
| Sampling settings (temperature, top-p, etc.) | Not supported | Supported |
| Forced tool choice, structured output | Not supported | Supported |
| Thinking text | Not shown | Shown |
| Usage | Counts against your Claude plan's limits | Billed in inference.sh credits |
Runs of the Claude Code models are free on inference.sh. Like any free app, the workspace still needs a balance or a saved payment method to start a run. See Billing: free runs.
Reconnecting
A token lasts one year. The Claude Code credential page shows the date it expires, with a warning in the last 30 days.
After that date the credential shows as expired and runs stop with "your Claude Code key expired on
To reconnect, run claude setup-token again and paste the new token in vaults → credentials → Claude Code. Saving it starts a new year.
Privacy
- The token is stored encrypted.
- It is used only for your own runs.
- It is never shared with other users or pooled.
Troubleshooting
| Problem | Fix |
|---|---|
| "your Claude Code key expired" | The token is a year old. Reconnect. |
| 401 | The token was revoked. Reconnect. |
| Runs fail after heavy use | You hit your Claude plan's usage limits. Wait for the limit to reset or upgrade your Claude plan. |