Ollama Model Not Found in OpenClaw

OpenClaw discovers models from the Ollama host you configured. A model installed on a different host, user environment, or container is not automatically visible.

Confirm the native endpoint

The Ollama provider base URL should look like this:

http://localhost:11434

Do not add /v1. OpenClaw uses Ollama’s native API, and the OpenAI-compatible path can break tool calling.

List the provider inventory

openclaw models list --provider ollama

Compare the exact tag with the one installed in Ollama. model:latest and another named tag are distinct references even when they share model weights.

Check network boundaries

Inside Docker, localhost is the container itself. A Windows Ollama host and a WSL2 Gateway may also require an explicit reachable address. Prove that the Gateway environment can reach the Ollama native API.

Check suitability metadata

Guided setup prefers an installed model when Ollama reports tool support and at least 16K context metadata. A model can be manually runnable yet omitted from the automatic recommendation path.

Expected result

openclaw models list --provider ollama prints the exact model reference, and a real tool-enabled test turn completes without raw tool JSON appearing as plain text.