Openclaw timeouts

[sturdy nymph]

Hello, i have a problem. It is when i type a prompt(example:create a web server and a personal blog website on this local machine) openclaw says its running and the timer is ticking. Model starts a response(usually like "I'll start installing apache2" or something like that) and everything is going perfect. Maybe after 2-3 mins the entire chat stops and its either of these two errors: "This response is taking longer than expected. Send another message to continue." or "run error: LLM request timed out.". For my model provider i use a local ollama instance throguh API thats on the same network as the openclaw server. The model in question was glm-4.7-flash(tho same problem is with qwen3-coder) I verified that the ollama server did not crash and that the API is working the entire time. Openclaw version:OpenClaw 2026.5.6 (c97b9f7) and ollama version: 0.21.2. My ollama instance has a 3090 so model being to large is not the problem. I tried putting openclaw and ollama on the same machine, but same outcome with same models. Thanks for any response in advance

[plush cipher]

This smells like the model request idle/request timeout, not Ollama crashing. OpenClaw’s full agent turn is much heavier than a direct “hi” API probe: it includes session context + tool schemas, and local Ollama models often stop streaming long enough that OpenClaw aborts even though Ollama itself is still alive.

First thing I’d try is provider-scoped timeout + keep-alive, not a global agent timeout:

{
  models: {
    providers: {
      ollama: {
        timeoutSeconds: 300,
        models: [
          {
            id: "glm-4.7-flash",
            name: "glm-4.7-flash",
            params: { keep_alive: "15m" }
          }

        ]
      }
    }
  }
}

Use the exact model id shown by:

openclaw models list --provider ollama
openclaw models status

Then test through OpenClaw, not just curl/Ollama:

openclaw infer model run --model ollama/glm-4.7-flash --prompt "Reply with exactly: ok" --json

If that passes but normal chat still times out, it’s probably the full tool surface/context being too much for the local model. Try lean mode:

{
  agents: {
    defaults: {
      experimental: { localModelLean: true }
    }
  }
}

If it still fails, test disabling tools for that Ollama model: