Connect a model
Each workspace has its own model providers, API keys and prices. Your operator may have set up a default. Check Model settings before you add your own.
Choose a provider
| Hosted provider | Your own model server | |
|---|---|---|
| Examples | OpenRouter, OpenAI, Anthropic, Google Gemini, AWS Bedrock, Groq, Together AI, DeepSeek, Mistral, Cheaper Inference | Any OpenAI-compatible server (such as vLLM or Ollama), or LM Studio |
| Where your code goes | To that provider | Only to your server |
| What you need | An account and an API key | A server the Gryndel server can reach over the network |
| Cost | Pay per use (some free models) | Your own hardware |
With a hosted provider, the code under review is sent to that company. Check its data retention and training policy, and your organisation's rules, before using it on private code.
The Gryndel server sends requests to the model, so the model must be reachable from that server. A model running on your own computer usually isn't. Ask your operator about a shared model server.
Add it
- Open Model settings and click Add provider.
- Choose the Type, give it a Name, and paste the API key or server address.
- Pick a Default model and click Test connection.
- To make it run your reviews, tick Default provider.
- Click Add provider.
Model settings explains every field: context windows, agents at once, per-step instructions and prices.
Enter prices
The Costs page can only show what reviews cost if it knows each model's price. OpenRouter, Cheaper Inference and Together AI publish theirs. For the others, enter prices under Prices per model on the provider.
Next: Your team's settings.