openclaw - 💡(How to fix) Fix [Feature]: Auto-discover local Ollama instance and pre-register models on startup [1 comments, 2 participants]

Summary

On startup, probe localhost:11434 for a running Ollama instance and automatically register detected models into the gateway's model registry.

Problem to solve

Users running OpenClaw with local models via Ollama must manually configure the provider URL, set credentials, and add each model by hand. This creates unnecessary friction for anyone who wants a local-first, privacy-preserving setup — which is arguably OpenClaw's core value proposition.

If OpenClaw is already running on the same machine as Ollama (or inside WSL2), it should be able to detect the service silently on startup and surface available models without any YAML/JSON editing.

Proposed solution

1. Startup probe: send a lightweight GET to http://localhost:11434/api/tags (and optionally a configurable list of ports/hosts ) during gateway init.
2. Auto-register provider: if a 200 response is received, create or update an ollama-local provider entry in the gateway state — no user action required.
3. Model sync: parse the response to extract pulled model names (e.g. llama3, qwen2.5, deepseek-r1) and add them to the selectable model list.
4. UI hint: surface local providers first in the model selector so users immediately see their local models at the top.

A configurable flag (e.g. ollama.autoDiscover: false) should allow users to opt out if they run Ollama on a non-default port or prefer manual setup.

Alternatives considered

No response

Impact

MateClaw v1.0.101 ships this exact feature: auto-detects a local Ollama instance on startup, pre-configures six popular models (Qwen3, Llama, DeepSeek, Gemma, Phi, Mistral), and sorts local providers first in the model management UI.

Release reference: https://github.com/matevip/mateclaw/releases/tag/v1.0.101

Evidence/examples

No response

Additional information

No response