Hello OpenClaw Community & Devs,
I'm encountering a persistent issue when trying to use local Ollama models for subagents with sessions_spawn.
Setup:
- OpenClaw Agent is running on a Hetzner server.
- Ollama (v0.17.7) is locally installed with
llama3.2:1bmodels. - OpenClaw is configured for the Google provider (
google/gemini-2.5-flash) as default. - Ollama endpoint and a dummy API key were configured via
openclaw config set providers.ollama.endpoint "http://127.0.0.1:11434"andopenclaw config set providers.ollama.apiKey "ollama".
Problem:
Every attempt to spawn a subagent using sessions_spawn(model='llama3.2:1b', provider='ollama', ...) results in the error:
FailoverError: Unknown model: google/llama3.2:1b
It appears OpenClaw is automatically prefixing "google/" to the Ollama model name, even when provider='ollama' is explicitly specified.
Attempts so far:
- Direct specification of the model name
llama3.2:1b. - Explicitly setting
provider='ollama'in thesessions_spawncall (current attempt). - Passing
OLLAMA_HOSTandOLLAMA_API_KEYasenvparameters tosessions_spawn.
All attempts yield the same "google/" prefix error.
Question for the Community/Devs:
Is there a specific configuration, environment variable, or different syntax to correctly instruct OpenClaw to use a local Ollama model for a sessions_spawn call without the "google/" prefix being added? Or is this an architectural limitation/bug in OpenClaw?
Any help or workarounds would be greatly appreciated, as this is blocking our ability to run worker bots locally and save on API limits.
Thank you!