Saved connections
Save a server’s address, backend profile, and authentication once. Voice, Audio file, Cleanup, and Text to speech then select any connection that supports their operation, each with its own task options.
Open Connections on the activity rail or Manage connections… in a task’s picker. Search by name, backend, or address, then select an entry to edit it. For your first server’s required values, use Connect a speech server.
Add a connection while setting up a task
Section titled “Add a connection while setting up a task”- Open the task’s Settings cog, then Transcription, Cleanup, or Speech. Choose Add connection… from its connection picker.
- Name the connection and enter its server details. Choose Backend, base URL, and authentication. The task you came from is already selected as a use.
- Review available workflows. Expand Available in … workflows only to enable other operations your deployment exposes. This does not select it for those tasks. Allow HTTP explicitly only for a trusted plaintext server.
- Choose Save and return. The connection is saved and selected for this task.
- Choose the model and model profile, then Save in the options sidebar. whisper.cpp uses its server-loaded model. Choose Done to close options.
A failed save leaves the previous connection active and keeps the draft. Correct the indicated error and retry. When leaving unsaved edits, Discard keeps the active connection unchanged; Keep editing returns to the draft.
Create a library entry without using it yet
Section titled “Create a library entry without using it yet”Open Connections → Add connection, fill in server details and supported uses, and set After saving → Save for later. Choose Save connection to create an inactive entry. To use it immediately, choose a workflow under After saving and use Save and return instead.
If Save is unavailable, check the message for a missing name, URL, or supported use. Existing entries use Save connection for edits and Cancel to leave.
What goes where?
Section titled “What goes where?”| In Connections | In workflow options |
|---|---|
| Name and supported uses | Active connection |
| Backend profile and base URL | Model, model profile, and language |
| Authentication and stored key | Cleanup preset and instructions |
| Allow HTTP | Voice, speed, and feature enable switches |
| Transcription health path and headers | Timeouts and model/backend options |
Built-in local connections
Section titled “Built-in local connections”Each configured local runtime automatically appears as a Built-in connection on supported computers. Install its runtime and model, start it, and select the connection from a compatible task. No separate URL, API key, or connection creation is needed.
| Change you want | Where to do it |
|---|---|
| Install/start/stop the local service | Local runtime, or quick runtime controls |
| Choose its downloaded/loaded model | Runtime model selection |
| Set language, realtime, or cleanup behavior | The task’s options |
The runtime owns its connection name, endpoint, model, and uses. Built-in entries cannot be duplicated or deleted in the connection editor and do not count toward manual limits. Stopped runtimes stay visible but unavailable; tasks do not switch servers automatically. Previously saved local aliases remain intact.
Reuse a server
Section titled “Reuse a server”Edit the connection, enable another supported use, and Save connection. Select it in the other workflow, then choose that task’s model. Check the backend guides before enabling an operation: a successful metadata check does not establish every capability.
Use separate connections when URLs, credentials, or backend profiles differ. Custom transcription headers and health paths apply only to Voice and Audio file, not cleanup or speech generation.
Switch, edit, duplicate, or delete
Section titled “Switch, edit, duplicate, or delete”| Action | Result |
|---|---|
| Choose an Active connection in task options | Switches immediately after resolving any draft; restores its last model/options for that task |
| Manage connections… / Edit connection | Opens the catalog / selected entry; All connections returns to the list |
| Use for… in the editor | Selects this server for a supported workflow and opens model settings; checkmarks show current uses |
| … → Duplicate | Creates an inactive copy with no remembered model preferences; later edits/key replacements affect only the copy |
| … → Delete | Removes an inactive entry after confirmation |
To delete an active entry or remove a use, first choose another connection or None selected in every affected task. When switching with draft settings, Save or Discard resolves them; Keep editing cancels the switch. An unconfigured connection needs a model, and Cleanup/Speech must be enabled again.
Use unique names. You can save up to 32 manual connections per use: Voice, Audio file, Cleanup, and Text to speech. A reusable entry counts toward each use it enables. Separate entries may share an address with different credentials. See remembered model settings.
Test and protect credentials
Section titled “Test and protect credentials”Check a saved server
Section titled “Check a saved server”Select an entry, expand Connection check, and choose Check connection. This checks that saved entry and its credential without selecting it or using unsaved edits. Configure gateway headers and health paths under Transcription connection options.
It reads health/model metadata, without audio or prompts. Open the task’s model picker and use Refresh models (or Check server for whisper.cpp) to assess that task’s selection. See connection-check results for metadata, authentication, model-list, and stale-result meanings.
Keep or replace a key
Section titled “Keep or replace a key”Saved keys stay in Windows Credential Manager or macOS Keychain and are never shown again. Leave the password field blank to keep a stored key, enter a replacement, or explicitly remove it. Leaving the editor or hiding the workspace clears an unsaved key draft.
A duplicate can use the same stored key. Replacing the duplicate’s key does not change the original, and deleting an entry does not remove a key still used by another connection.
A successful check does not guarantee that inference will work or that the key has permission to run the selected model. See credentials and transport.
Discover speech voices
Section titled “Discover speech voices”Speaches, Kokoro-FastAPI, and vLLM-Omni offer Refresh voices under Text to speech → Speech. Search or type a voice ID. Speaches may identify voices for the selected model; server-wide lists are labelled. Refreshing neither selects a voice nor generates audio. Save to remember the voice for that connection/model.