Play with ollama localy just to try NO GPU
✅Solved
Hello let's see if if we can play openclaw with cpu only
Solution
Yep — OpenClaw itself is lightweight, so CPU-only is totally fine. The hard part is the LLM inference, and on CPU it’ll be slow (sometimes very slow), but it’s doable for “playing around”.
1) Install Ollama and start it (it usually runs a local server on
2) Pull a model you want inside Ollama (pick something that fits your RAM).
3) Opt OpenClaw into Ollama discovery (Ollama doesn’t need a real key; any string works):
- set an env var:
- (then OpenClaw can auto-discover Ollama models if you don’t define a full
4) In OpenClaw, verify it sees Ollama:
-
5) Set your agent’s default model to an Ollama one (format is
- Local models overview: https://docs.openclaw.ai/gateway/local-models
Easiest path: Ollama (local, CPU-only)
Docs: https://docs.openclaw.ai/providers/ollama1) Install Ollama and start it (it usually runs a local server on
127.0.0.1:114342) Pull a model you want inside Ollama (pick something that fits your RAM).
3) Opt OpenClaw into Ollama discovery (Ollama doesn’t need a real key; any string works):
- set an env var:
OLLAMA_API_KEY="ollama-local"- (then OpenClaw can auto-discover Ollama models if you don’t define a full
models.providers.ollama4) In OpenClaw, verify it sees Ollama:
-
openclaw models list5) Set your agent’s default model to an Ollama one (format is
ollama/<model>/modelReality check (important on CPU)
The FAQ is pretty blunt: local/small/quantized models are often not great for real assistant + tools, and they’re more prompt-injection-prone. If you keep this bot connected to real tools/surfaces, consider sandboxing + strict allowlists.- Local models overview: https://docs.openclaw.ai/gateway/local-models
The AI that actually does things. Emails, calendar, home automation - all from your favorite chat app. New shell, ready to help. The lobster way.
83,815Members
Resources
Recent Announcements
Similar Threads
Was this page helpful?
