Choosing models
Every coding session uses a model you chose. Switch providers between sessions, or mid-session with the context limits stated.
Available today
Section titled “Available today”| Provider | How you connect |
|---|---|
| Anthropic | your API key |
| OpenAI | your API key |
| Google Gemini | your API key |
| Any OpenAI-compatible endpoint | its address and your key |
| Ollama | detected on your machine |
| LM Studio | detected on your machine |
OpenRouter and llama.cpp-style servers work through the OpenAI-compatible option with their own address. See the integrations catalog for every entry and its status.
Local models
Section titled “Local models”Vekrun tells you two things separately: whether a runtime is installed, and whether it is running. A local model handles explanations and smaller coding tasks entirely offline. Larger multi-agent work may need stronger hardware or a hosted provider, and Vekrun says so rather than failing quietly.
What Vekrun promises
Section titled “What Vekrun promises”- Your key never shows in logs or in the interface, and it is never handed to tools or remote machines.
- Vekrun never switches to a different provider on its own. Private data does not go anywhere you did not choose.
- Where a provider does not publish a model’s context size or price, Vekrun shows “unknown” instead of guessing.
- A failed request is retried, and you can see the retry in the session.