Stellaris Companion uses one selected AI provider and model for both the built-in Advisor and Chronicle. Gemini is the default, but it is not required.
On first launch, you can add a Gemini key or skip that step. To use another provider, open Settings → Intelligence Uplink after onboarding.
Compare the choices
| Provider | Where inference runs | Credential | Before connecting |
|---|---|---|---|
| Gemini | Gemini API key | Create a key in Google AI Studio. | |
| Ollama | Your computer by default; cloud models are remote | None for the local server | Start Ollama and load a suitable model. |
| LM Studio | Your computer | None for the local server | Start its local server and load a model. |
| OpenRouter | Remote provider selected through OpenRouter | OpenRouter API key | Confirm that the account has usable credits or quota. |
| Custom endpoint | Wherever the configured server runs | Optional | Use an OpenAI-compatible chat-completions server. |
Use a local provider when keeping model inference on your computer matters most. Use Gemini or OpenRouter when you prefer a hosted model without managing local model hardware. A custom endpoint is best for an existing compatible service or private-network server.
Configure and check a provider
- Open Settings → Intelligence Uplink.
- Choose the provider.
- Enter its credential or server URL when required.
- Select Check connection to find available models.
- Choose a model, then select Test model.
- Save the configuration.
The connection and model tests do not send campaign data. A complete test reports one of three outcomes:
- Advisor + Chronicle ready: the model supports the preferred structured response.
- Advisor + Chronicle ready — compatibility mode: the model works through a compatible fallback.
- Advisor ready — Chronicle not compatible: use this model for questions, but choose another before generating Chronicle prose.
Full campaign briefings generally need an active context window of at least 32K tokens. For Ollama and LM Studio, the context configured by the local server matters—not only the model’s advertised maximum.
Connection safety
Ollama defaults to http://127.0.0.1:11434/v1, and LM Studio defaults to http://127.0.0.1:1234/v1. Public custom endpoints must use HTTPS. Plain HTTP is accepted only for localhost and private-network model servers.
Advisor questions, Chronicle instructions, and extracted campaign context are sent to the selected model. Raw .sav files remain local. Ollama cloud models process requests remotely even though Ollama is the selected provider.
If the check fails
- Provider unavailable or timed out: start the local server, confirm its address, and allow time for the model to load.
- Authentication or billing failed: replace the key or check the remote provider’s quota and credits.
- Model not found: check the connection again and choose a model currently returned by the server.
- Context limit: increase the active context to at least 32K or choose a larger-context model.
- Invalid response: confirm that a custom endpoint supports OpenAI-compatible chat completions.
- Chronicle not compatible: keep the model for Advisor questions or select a model that passes the structured-output test.
Once the provider is ready, connect your first save.