GranularConnect other AI providers
Use DeepSeek, Kimi, GLM, Grok, Groq, Mistral, OpenRouter and more in Granular. Pick the provider, paste your API key, and Granular fills in the models.
Updated 2026-10-20
Granular can use models from many AI companies with your own API key: DeepSeek, Kimi, GLM (Z.ai), xAI (Grok), Groq, Mistral, Together AI, Fireworks AI, Cerebras, OpenRouter and OpenCode Zen. Pick the provider from a list and paste your key. Granular checks the key and fills in the models for you.
You pay the provider directly at its own prices. This doesn't use Granular's AI credits.
An API key is a long password from the provider's website that lets an app use your account there.
Before you start
- An account and an API key with the provider. Most providers need some credit on the account before the key works.
- Granular, with any plan.
Add a provider
- In the row of controls under the message box, click the button that shows your current connection's name, such as Claude or G.hosted.
- Click Add your AI agent, then, under Keys & providers, click Custom provider (its tag says Add).
- Pick your provider from the list. ("Pick your provider - we fill in the details, you paste a key.")
- Paste your key in API key. If you don't have one yet, click Get a … key › (or … API docs ›) to open the provider's page.
- Click Find models with my key, or just click Add & test: Granular looks up the models first either way.
What you'll see
When your key works, Granular fills Model ids with up to six current models and says so, for example "Key works. We picked 6 current models - 40 more are available; add any by its id." The first model is the default.
Then, as with any custom provider, Add & test sends one tiny message to check everything works, and the provider joins your Accounts list. The session switches to it. See Add a custom AI provider for the test messages.
To pick a model for a message, use the model button under the message box. It lists the models you added first, then the provider's full current list, grouped and searchable, so you can try any of them without editing the provider. Granular keeps that list up to date by itself. See Pick and pin models.
OpenRouter in one click: instead of pasting a key, you can click OpenRouter (tag Sign in) under Add your AI agent and approve in your browser. See Sign in with OpenRouter.
Used by OpenCode and Pi too: a provider key you add here is also used automatically by OpenCode and Pi, if you use them, with no extra setup inside those apps.
The providers
| Provider | What it offers |
|---|---|
| OpenRouter | One key for hundreds of models from many companies |
| DeepSeek | DeepSeek's own models |
| Kimi (Moonshot) | Kimi models |
| GLM (Z.ai) | GLM models with a pay-as-you-go key |
| xAI (Grok) | Grok models |
| Groq | Fast open models |
| Mistral | Mistral's own models |
| Together AI | Many open models |
| Fireworks AI | Fast open models |
| Cerebras | Fast inference |
| OpenCode Zen | Models tested by the OpenCode team |
A few things to know:
- OpenRouter: Granular only offers models that can use tools, reply in text and haven't been retired. Agents need tools to read files and run commands.
- GLM (Z.ai): use a pay-as-you-go key. The GLM Coding Plan's terms limit it to tools Z.ai lists, so it isn't offered here.
- OpenCode Zen: its Claude, GPT and Gemini models use other addresses, so they aren't part of this connection.
- For every provider, models that can't chat (embeddings, speech, images and so on) are left out.
Change or undo it
- Add more models: edit the provider (its ⋯ button in Accounts, then Edit provider) and add model names to Model ids, separated by commas. Up to 20.
- See or change the filled-in details: click Edit details on the summary card before saving.
- A provider that isn't in the list: click Another provider - type its details and fill in the form yourself. See Add a custom AI provider.
- Remove it: its ⋯ button, then Remove.
Limits
- Your key, your bill. Usage is charged by the provider. Granular's AI Add-on credits aren't used.
- Six models to start in Model ids, up to 20 per provider. The model button also offers the provider's whole current list, and any model you pick there is kept, even if the provider later drops it (it's then marked Unavailable).
- Providers on the OpenAI-compatible format go through Granular's built-in translation, which drops the model's "thinking" steps and prompt caching. See Add a custom AI provider.
- Your key is stored encrypted on this computer. It isn't sent to Granular, and it isn't listed in the Vault.
Troubleshooting
If you see "… didn't accept that key. Check you copied all of it and that it's still active."
Copy the key again from the provider's website, with no spaces at either end.
If you see "Your … account has no credit left. Add credit on …'s website, then try again."
Add credit to your account with the provider.
If you see "… is asking us to slow down. Wait about a minute, then try again."
The provider is limiting requests. Wait, then click Find models with my key again.
If you see "… is having problems right now. Try again in a few minutes."
The provider is down. Try later.
If you see "Couldn't reach …. Check this computer is online, then try again."
Check your internet connection. On a company network, the provider's address may be blocked.
If you see "Your key works, but … didn't list any chat models we can use - type a model id instead."
Type a model name from the provider's website in Model ids.
If you see "… couldn't list its models just now. You can still type model names yourself."
Type the model names from the provider's website, then click Add & test.