Multi-Provider Chat

Sythoria sends model requests from its Rust backend to the endpoint saved on the selected model. It does not route prompts through a Sythoria service.

For a hosted model, the provider receives the request and credential. For a local model, the request goes to the local endpoint you configured. Search, MCP, transcription, updates, and artifact networking have their own network paths.

Provider presets#

These values come from v0.4.4. They are editable defaults, not a guarantee that the provider currently offers the model to your account.

Add a model#

1. Open Settings > AI Models & API Keys.
2. Select Add Model.
3. Choose a preset or Custom.
4. Check the API base, request provider, and exact model ID.
5. Enter a credential if the endpoint requires one.
6. Save and run the connection check.

Credentials are encrypted in secrets.enc. The root encryption key is kept in the operating system credential vault.

Switching models#

The conversation stores a model selection. Changing it affects later requests, and the new provider can receive the history included with those requests. Start a new conversation when earlier messages should not cross provider boundaries.

Comparison columns and subagents capture their own model and project context. Changing the global selector later does not rewrite a run that is already active.

Custom model endpoints#

Custom models use an OpenAI-compatible chat-completions URL. Enter the full endpoint expected by the server. Use HTTPS for remote hosts and grant local-network access only for an exact local service you trust.

Do not send a production credential to an endpoint whose operator or TLS setup you have not verified.