Skip to content

Network connections

Connection When it occurs Data boundary
Local Ollama loopback service Local summaries, chat, local embeddings, model management, and global-index background embeddings. The bundled runtime listens on a loopback address. Synthezia prefers 127.0.0.1:11435 and can select another local loopback host when needed.
Whisper model download You start a download or opt into background model preparation. The selected model identifier and model download are sent to the Hugging Face source used by the application.
Ollama model pull You start a pull or opt into background model preparation. Synthezia asks the local Ollama runtime to pull a named model. The runtime’s configured registry determines the upstream destination.
OpenAI-compatible generation API You select External API and generate a summary or chat answer. The configured provider receives generation context as described in remote provider data flow.
OpenAI-compatible embedding API External API is selected for provider-aware embeddings or a global-chat query embedding. The configured provider receives embedding input such as transcript chunks, summaries, or the current query.

What does not require an external AI endpoint

Section titled “What does not require an external AI endpoint”

The current imported-media and live-recording conversion path uses local FFmpeg, and transcription uses local Whisper. Local Mode sends processing requests only to the loopback Ollama runtime after models are present.

The inspected application runtime does not register a user-facing automatic application-update or release-check request. This does not describe network activity from macOS, development tools, a remote provider, or the Ollama runtime itself.

Before selecting External API, review the endpoint’s domain, transport configuration, data handling, retention, and access controls. Synthezia sends requests to the base URL you configure and does not provide a provider-neutral retention guarantee.

  • A missing local model needs a network connection before Local Mode can complete the feature that requires it.
  • A local Ollama startup failure prevents the loopback service from answering local summary, chat, or embedding requests.
  • An external provider can reject a model name, chat request, embedding request, authorization credential, or connection.