OpenClaw with OpenRouter

OpenRouter is the answer to a specific wish: one API key that reaches models from many vendors, with a router that picks for you when you do not want to choose. OpenClaw talks to it over the same OpenAI-compatible transport it uses for other proxy providers, which keeps the setup short. Here are the two ways to sign in, how model references are written and resolved, and the three media tools the same key can back.

OpenRouter routes requests to many models behind one API and one key.

Two ways to sign in

• OAuth: send the login command for OpenRouter in a private chat, or run onboarding with the OpenRouter OAuth choice. OpenClaw opens the browser sign-in, exchanges the code for an OpenRouter API key and stores it in the default auth profile; on a headless host it prints the URL and asks you to paste the redirect. From chat, the browser callback uses the Gateway’s managed Tailscale HTTPS address, so the browser has to be on the tailnet.
• API key: create one on the OpenRouter keys page and run onboarding with the API-key choice, or put the variable in the config’s environment block.
• Login saves access without choosing a starter model; onboarding defaults to the auto router, and the Control UI has the same credential-only flow under Settings, Models, Connect, with the model menu for the actual choice.

Model references

A reference is written as openrouter, then the upstream provider, then the model. Three bundled starters, the auto router and two Kimi models, enrich a nonempty public catalog; everything else resolves dynamically against OpenRouter’s live catalog, including the fusion router. A failed live request reports a discovery failure rather than substituting the bundled rows, and a successful empty response stays empty.

Beyond chat

• Image generation: an OpenRouter image model under the media models block backs the image tool through the dedicated images API; Gemini image models also receive aspect ratio and resolution hints.
• Video generation through the asynchronous videos API: the bundled default supports four, six or eight seconds, 720p or 1080p, and 16:9 or 9:16; video-to-video is not supported because the upstream API only accepts text and image references.
• Music generation through chat-completions audio output, with its own model and timeout under the same block.

OpenClaw models and providers explains how OpenClaw resolves a reference and fails over between providers, and What OpenClaw costs is where the router’s per-request pricing belongs in the monthly picture.

When it fits

One key when you want to try models across vendors without five accounts, auto routing when you would rather not choose, and the markup is the price of both. The docs’ model advice still applies through a router: the strongest model you can afford for any agent with tools, and cheaper ones only for low-stakes chat. OpenClaw and NVIDIA is the free-open-models alternative for the same one-key convenience.

On Diali

On Diali credits are included and the assistant ships with a managed model list, so a router is something you add only if you want models outside it. Hosted OpenClaw on Diali is the assistant and Diali pricing shows what the credits cover.

• OAuth or an API key; both land in an auth profile.
• openrouter, provider, model; auto when you do not want to pick.
• The same key can back images, video and music.