Pick any OpenRouter model per workflow step
Point AI agent steps at any OpenRouter model with one API key: paste a model id per step, pay OpenRouter directly, and keep the rest unchanged.
- Category
- AI models
- Connect
- Paste an API key · managing connections
- Operations
- bring your own key
- Triggers
- none yet
OpenRouter is one key in front of hundreds of models, and MadMax treats it that way: a provider connection for AI agent steps, not an app to automate. Connect your key, bind it on an agent step, and type the model id for that step. One workflow can run a cheap model for classification and a stronger model for the final draft, each on its own step, with an approval between the draft and the send if the output has consequences. Runs show each step's prompt, output, tool calls and token usage, so comparing models on real work is reading the run history. Describe the workflow in plain English, let MadMax wire it, then swap in your OpenRouter model where you want it.
What teams build
- A helpdesk webhook is classified on a small, fast model, a switch routes refund requests to a stronger model that drafts the reply from a knowledge base, and an approval in Slack gates the send.
- A nightly schedule pulls unlabeled tickets from a store, and a loop runs two agent steps, each bound to a different model id, on the same ticket and saves both answers for a morning comparison email.
- A file upload trigger feeds an extract step bound to a
:freemodel; an if step checks the required fields came back and routes misses to a send email asking for a re-upload.
How it connects
Create a key at openrouter.ai and paste it (sk-or-v1-…). The connection is only the key: on each agent step set Provider to it and type the model id verbatim as openrouter.ai lists it (openai/gpt-5.6-luna-pro), routing suffixes like :nitro or :free included. OpenRouter's catalog changes weekly, so there is no picker list to go stale. Bound steps run on your OpenRouter account and MadMax does not meter that spend; unbound steps and the canvas chat stay on the platform model. The live web-search tool is platform-model only, while browser and code tools still run. See managing connections and AI settings and models.
Related
Works with
Describe the workflow. MadMax builds it and tests it on your data.
Free to start, no card. Connect your apps, say what should happen in plain English, and watch each step run against real data before anything goes live.