AI settings
Open Settings → App → AI to select the language-model route used for text processing. Echo retrieves the current catalogue at runtime, so provider and model names can change without this page becoming a permanent catalogue.
Provider profiles
The panel separates profiles by transport:
| Section | Request path |
|---|---|
| Cloud Models | Echo sends the prompt and relevant workflow text through the authenticated Budgie gateway to the selected provider. |
| Local Models | On-device entries and detected Ollama, LM Studio, or vLLM models run through a local endpoint. Echo scans the standard loopback endpoints in the background at startup; use Rescan after starting or changing a local server. |
| Custom Endpoints | Echo sends the request directly to the endpoint URL you configure. The endpoint's operator controls its storage, logging, and retention. |
Selecting a cloud profile clears the active custom profile, and selecting a custom or detected local profile clears the active cloud profile. A detected local model is registered behind the scenes as a custom-compatible endpoint, but remains in the Local Models section. Discovery is available only to accounts with custom-endpoint access and probes loopback addresses only; it does not scan the local network.
If a detected local server requires authentication, choose Set API key beside its name. Echo validates the credential, stores it encrypted on the device, and rescans automatically. The key is then reused for that provider's discovered models without being exposed in Settings, logs, or analytics.
Custom endpoints require plan access. Supply an API key only when that endpoint requires one. Secret values are not reference material: never copy API keys into documentation, workflow fields, screenshots, diagnostics, or support reports.
General settings
Default temperature controls the balance between consistency and variation for LLM calls. Echo defaults to 0.1. One-time prompts and newly added LLM recipe nodes start from this value, and their sliders can override it for an individual run or node. Existing recipe nodes without an override continue to inherit the current default. Providers that do not expose temperature ignore this setting.
Privacy mode
Privacy mode (local-only) blocks gateway routes and clears an active cloud selection. It also blocks a Custom Endpoint when its URL is a non-local custom endpoint. An on-device model or localhost custom endpoint remains allowed.
Privacy mode does not change the behaviour of HTTP, download, shell, plugin, or recipe actions. Review those destinations separately; the toggle is an inference-route guard, not a whole-application offline mode.
Where text can leave the computer
Cloud and remote custom profiles receive the prompt plus any transcription, selected text, constants, or earlier workflow outputs interpolated into it. Audio is handled by the transcription route; an LLM request normally receives text, not the original audio.
Before selecting a remote profile, review the workflow's inputs and the provider's own terms. For the rest of Echo's network boundaries, see Privacy & Data.
Troubleshooting a profile
- Check that the intended row is selected and that Privacy mode is not disabling it.
- For a local server, start the server, load a model, choose Rescan, and use Set API key if Echo reports that the server is protected.
- For a custom endpoint, verify its URL, model identifier, provider type, and required credential using the built-in test action.
- For a cloud route, confirm sign-in and available quota under About.
- Inspect Events, then Logs, and redact values before sharing them.