Providers
OpenRouter
Hundreds of models, one key
Connect it
- Open Settings → Providers and find OpenRouter.
- Press Connect.
- Sign in to OpenRouter in the page that opens, and it makes a key for Conch. Nothing to copy. Or paste a key you already have: it starts with
sk-or-.
Conch checks the key before keeping it.
What you get
Models from every lab in one list (Claude, GPT, Gemini, Llama and more), each with its price beside it. You pay OpenRouter as you go.
Conch runs the conversation itself: it keeps the thread, holds the connections to your apps and hands the model their tools, and asks you before anything changes. So apps, memory, skills and the browser all work, with models that can call tools.
Good to know
- Conch supplies the tools. Tool-capable models can use memory, connected apps and files in this conversation’s work folder, with the same permission checks and Undo tracking. Commands require the OS sandbox, cannot access the network and have no unrestricted fallback. Models that weren't made for tools are labelled Chat only in the picker; if you use one anyway, Conch lists the tools in its instructions so it can still try. This works for any model OpenRouter has, not only the ones the picker shows.
- Spend is tracked. Conch records what each turn cost, and you can set a budget. See Offline and at a limit.
- What a model already read costs less. Claude and Gemini models are asked to keep the instructions and the chat in their cache, and the others keep it by themselves, so each step of a long job pays full price only for what's new.
- A busy model is waited for. When OpenRouter says to slow down, Conch tries again a few seconds later by itself, then hands the chat to your fallback if it's still busy.
- A key with an end date says when it runs out on its card, from two weeks before. Once it's past, OpenRouter no longer knows the key: make a new one at openrouter.ai and press Add a key.
- Pictures work with models that can see. A file card warns you when the chosen model can't use it.