openclaw - 💡(How to fix) Fix [Bug]: idletimeout seconds [1 comments, 2 participants]

## Steps to Reproduce

1. Configure LM Studio as a local OpenAI-compatible provider with a large model that takes longer than 60–120 seconds to cold load.
2. Set `models.providers.<id>.timeoutSeconds` to a high value, such as `900`.
3. Ensure the model is unloaded/cold in LM Studio.
4. Run an OpenClaw agent request against that provider/model.
5. Observe whether OpenClaw waits for the configured provider timeout or fails/fallbacks before the first token.
6. Compare this behavior to the former `agents.defaults.llm.idleTimeoutSeconds` setting, which previously prevented premature failover during cold load.
7. Run `openclaw doctor --fix` and observe that the old `agents.defaults.llm` block is removed as legacy.

    Compare against the old legacy setting, which previously allowed the same cold model to load successfully:

{
  "agents": {
    "defaults": {
      "llm": {
        "idleTimeoutSeconds": 300
      }
    }
  }
}

### Expected behavior

## Expected Behavior

When `models.providers.<id>.timeoutSeconds` is configured for a local LM Studio / OpenAI-compatible provider, OpenClaw should apply that timeout to the entire model request path, including:

- provider HTTP request timeout
- OpenAI-compatible streaming request
- time-to-first-token wait
- model stream idle watchdog
- embedded-runner failover timing

For example, with:

```json
"models": {
  "providers": {
    "local-workers": {
      "baseUrl": "http://(agent):1234/v1",
      "auth": "token",
      "api": "openai-completions",
      "timeoutSeconds": 900
    }
  }
}

### Actual behavior

```md
## Actual Behavior

In OpenClaw `2026.4.26`, `openclaw doctor --fix` removes the previous `agents.defaults.llm.idleTimeoutSeconds` setting as legacy and reports:

```text
Removed agents.defaults.llm; model idle timeout now follows models.providers.<id>.timeoutSeconds.

### OpenClaw version

2026.4.26

### Operating system

macos 26.4.1

### Install method

Installed via pnpm / npm global install on macOS.

### Model

Qwen3.6-35B

### Provider / routing chain

OpenClaw gateway → local-workers provider → LM Studio 

### Additional provider/model setup details

_No response_

### Logs, screenshots, and evidence

```shell

Unknown. Last working behavior was prior to 2026.4.26, when `agents.defaults.llm.idleTimeoutSeconds` was still accepted and effective.