#helper
fading tartanBOT
foggy cryptBOT
You can have both Ollama models installed at the same time — the main question is how you want to switch between them inside OpenClaw.
1) Install both models in Ollama
Use the exact model IDs Ollama expects (check with ollama list):
ollama pull qwen2.5:7b-instruct
ollama pull qwen2.5:14b-instruct
ollama list
2) Enable the Ollama provider in OpenClaw
Simplest (Ollama doesn’t need a real key, any value works):
export OLLAMA_API_KEY="ollama-local"
or:
openclaw config set models.providers.ollama.apiKey "ollama-local"
Then verify OpenClaw can see your Ollama models:
openclaw models list --provider ollama
Note: OpenClaw’s auto-discovery only keeps models that report tools support. If your Qwen “instruct” model doesn’t show up, that’s usually why. (Docs: https://docs.openclaw.ai/providers/ollama)
3) Pick how to use “easy” vs “advanced”
Option A (most common): switch on-demand with /model
In Discord chat you can do:
/model→ pick from the list- or directly:
/model ollama/<model-id>
So you keep 7B for normal stuff, and switch to 14B when needed.
Option B: create aliases (so it’s 1 word to switch)
Add aliases:
openclaw models aliases add easy ollama/qwen2.5:7b-instruct
openclaw models aliases add advanced ollama/qwen2.5:14b-instruct
Then in chat:
/model easy/model advanced
Option C: two agents (fast agent + deep agent)
If you want a “fast bot” and a “smart bot” with different defaults, run two agents and route tasks to the right one (Docs: https://docs.openclaw.ai/concepts/multi-agent). This is heavier setup, but nice if you want separation.
Docs pointer for “different models for different tasks”:
https://docs.openclaw.ai/help/faq#how-can-i-use-different-models-for-different-tasks
If you paste your openclaw models list --provider ollama output, I can tell you the exact /model ... strings to use (matching your local model IDs).
thorny panther