* Download a GGUF model from a HuggingFace URL with progress reporting * @param url Full URL to the GGUF file (e.g., https://huggingface.co/user/re

* Fetch models from all sources (remote servers + refresh local cache)

* Get logs filtered by criteria

* Get the latest base64 frame directly (for pre-processor). * Computes base64 lazily from raw bytes when needed.

Get the current sampler parameters

* Check if a server string represents a local model

Check if a model is loaded

Check if phone number (or key) is whitelisted for messaging. Uses Redis so all workers share the same state. TTL-based expiry is handled

* Logical-model view of the registry. * * One entry per non-mmproj file (model). Each entry pairs that file with its * assigned projector (if

* Remove an inference server address

Save tokens to file

* Send messages to a model by name. Routes local models directly, remote models via fetchResponse.

Set the sampler parameters

Set whether to use GPU acceleration (Metal) Must be called before load_model to take effect

* Teardown a specific master stream

Unload the current model and free memory

Updates a user's app_metadata in Auth0 to set their subscription status and link their payment provider IDs (Stripe or Apple). To CLEAR

* Add an inference server address

Add phone number to appropriate whitelist with 24h expiry in Redis. Optionally map a user-chosen key (like "pizza" or "Hey, I'm Roy") to the

* Capture a frame from a video stream. Returns base64 PNG or null. * * For Tauri screen capture: uses raw frame bytes from the channel directly.

Clear the audio channel

Clear the video channel

Ensure the unified capture is running

Find an mmproj file for a given model file

Hide the selector window and complete the awaiting request with a selection.

* Get the active service for a stream type without acquiring it. * Returns undefined if no service is active.

* Get or create a subscriber for an agent and stream type. * The subscriber accumulates transcripts independently for each agent.

Return the list of models supported by this handler.

Get existing Stripe customer ID or create a new one with the Auth0 email. This ensures the Stripe customer always has the Auth0 email, preve

Returns (is_pro, is_max, is_plus) for a given Apple product_id.

Single source-of-truth check for active subscriptions across all providers. - Apple: queries Auth0 metadata by email (cache of client-verifi

* Subscribe to model list changes

Prompt for the secret/config value for a given channel using TUI

Resolve a key or phone number to an actual whitelisted phone number. First checks if the input is a key in the Redis key_to_phone mapping.

Set interim text (replaces previous interim, for streaming services)

Set the App Group container path (called from Swift/iOS side)

Start video capture stream If capture is already running (for audio), reuses the existing stream

Verify and decode a JWS signed by Apple (StoreKit 2 transaction or notification). Apple StoreKit 2 uses certificate chain validation (x5c he

Convert OpenAI format messages to Gemini API format. Returns: (system_instruction_dict or None, contents_list)

Convert OpenAI format messages to Gemini API format. Returns: (system_instruction_dict or None, contents_list)

A pure, simple proxy that streams requests and responses directly.

* Cancel an ongoing download - signals Rust to stop and delete partial file