Hive can run fully offline with Ollama, use provider APIs directly, or route model traffic through LiteLLM. Each agent and Staff profile can prefer a different model.
Local models with Ollama
- Install Ollama.
- Start it with
ollama serve. - In Hive, open Models → Local.
- Pull a model, wait for the progress indicator, and run the connection check.
- Select the model when creating an agent or editing a Staff profile.
The default Ollama endpoint is http://127.0.0.1:11434. An unprefixed model ID is treated as an Ollama model. The setup check only requires a reachable Ollama service; you can pull a model afterward.
Direct cloud providers
Configure provider keys in Models or at launch:
| Provider | Environment variable | Model prefix |
|---|---|---|
| Anthropic | ANTHROPIC_API_KEY |
anthropic/ |
| OpenAI | OPENAI_API_KEY |
openai/ |
| Gemini | GEMINI_API_KEY |
gemini/ |
Environment values take precedence over database values. Hive masks stored credentials in the UI and logs. Use an env:VARIABLE_NAME reference when you want configuration to name a secret without storing its value.
LiteLLM gateway
The optional gateway centralizes provider keys, routing, failover, and spend controls. Set:
LLM_GATEWAY_URL=http://127.0.0.1:4000
LLM_GATEWAY_KEY=your-revocable-key
Gateway model aliases use the gateway/ prefix, such as gateway/hive-smart or gateway/hive-coding. Keep the gateway bound to localhost unless you deliberately secure a remote deployment.
Choosing models
- Use a small local model for classification, formatting, and frequent scheduled work.
- Use a strong reasoning or coding model for Colony roles that make consequential decisions.
- Set a Staff model preference for the normal default; a Colony launch model plan can override it for that run.
- Test tool calling before assigning a model to an agent that depends on MCP or sandbox tools.
If a provider fails, see Troubleshooting.