| # OpenClaw | |
| *Least verified page of this folder.* What I read in the OpenClaw documentation index (2026-10-15): providers are | |
| configured under `models.providers` (also in the Control UI, Settings → Models), a custom base URL is supported and | |
| LiteLLM is listed among local proxies, models are referenced as `provider/model`, and the CLI has | |
| `openclaw models set <provider/model>`. I did not read the page with the exact key names. | |
| Plan: | |
| 1. Run LiteLLM with the hook of `02-litellm.md` (A). | |
| 2. In OpenClaw add a custom provider whose base URL is `http://<host>:4000/v1` and key is a LiteLLM key, and select | |
| the model `s1d1-auto` (follow the *Custom providers* page for the exact keys). | |
| 3. If OpenClaw talks to several chat apps, give each channel the same model name: the router sees every first message. | |
| Open question: whether OpenClaw's plugin SDK offers a pre-model hook; routing at the gateway does not need it. | |