Check if a provider ID is a built-in provider

Returns the default base URL for a given provider ID.

Refreshes the cached provider items. Call this when providers or connection status changes.

Remove a hidden base prompt prefix for a given mode if it was persisted previously.

Returns the default models for a given provider ID. This is used when the user has not added any custom models for that provider.

Transcribe audio samples - Parameter samples: 16kHz mono PCM float samples - Returns: Transcription result with text and confidence

Download/prepare models for transcription.

Get reasoning config for a specific model, with smart defaults for known models

Refreshes the HAL liveness snapshot. Call only from an existing background hardware-refresh path; UI and capture selection consume the cached value.

Merge consecutive turns from the same speaker separated by short gaps. Fewer, longer turns give the ASR more context per slice and read better. Merged

Schedule opening the gate once. Safe to call multiple times.

Cancel an in-flight prewarm. Called on abort / new recording start — NOT on a normal stop, because AI post-processing runs after stop and benefits fro

Check if models exist on disk without loading them (async). This method performs an accurate async check for providers that require it (e.g., `AppleS

Global check if a model is a reasoning model (requires special params/max_completion_tokens)

Reset the service state

Route recovery and engine retry paths use this completion barrier so the old AVAudioEngine and its AVAudioIOUnit are fully deallocated before a replac

Waits for every release already submitted to this drain. This is used at capture start so a fire-and-forget retirement from a non-route path cannot ov

Await until the gate is open. Returns immediately if already open.

Applies custom dictionary replacements to transcribed text. Replaces trigger words/phrases with their designated replacements. Uses case-insensitive m

Hide expanded command output notch - force close regardless of hover state

Invalidates the dictionary cache. Called when settings change.

Returns true if the selected AI provider is currently verified/configured, regardless of the AI toggle or prompt selection. Used to gate prompt-mode h

Whether the model rejects the `temperature` parameter. Covers reasoning models plus Anthropic models that have deprecated temperature (Opus 4.7+, Sonn