Run OpenClaw with a local Ollama model

Local setup sequence

1. 1

   Install and pull a model

   Choose a model whose hardware, context, and tool support match the workload.

   ollama pull gemma4
   ollama list

   Copy
2. 2

   Run OpenClaw onboarding

   Select Ollama and Local only or Cloud + Local.

   openclaw onboard

   Copy
3. 3

   Verify discovery

   OpenClaw queries the native Ollama catalog and capability metadata.

   openclaw models list --provider ollama

   Copy
4. 4

   Select and test

   Set the exact discovered reference, validate, and use a narrow smoke test.

   openclaw models set ollama/gemma4
   openclaw config validate

   Copy

Common boundary mistakes

• Do not configure `http://host:11434/v1`; omit `/v1` for the native Ollama provider.
• In Docker, use a host-reachable address such as `host.docker.internal`, not container loopback.
• Local models may report zero API rate, but hardware, power, latency, and operator time remain costs.
• A model that answers plain text may still fail an agent turn if tool calling or context capability is insufficient.