Skip to content

AI assistant

reMaestro can point at an AI model to untangle a stuck activity, turn a sentence into a working rule, or build a remote from a photo. It is completely optional and off by default.

  • Bring your own key and provider. Choose Anthropic or any OpenAI-compatible endpoint (or a mock provider for offline). Nothing is sent anywhere until you set a key.
  • Your prompts go where you point them. Reasoning and vision talk to the provider you configured and nothing else — there is no reMaestro cloud in that path.
  • Separate models for reasoning (fixing states, building rules) and vision (a remote from a photo), so you can spend where it matters.
  • Speech is kept separate on purpose: a local Whisper endpoint can transcribe without any audio leaving the house, and a blank local URL will not silently fall back to the cloud.
  • Transcription is the one place a reMaestro-hosted service can sit in the path, and only if you tick cloud speech in Settings → AI. It is off unless you turn it on, and it moves audio only — your prompts still go straight to your own provider.

Everything else in reMaestro works with no AI at all.

Out of the box, nothing anybody said — a conversation lives a few minutes in memory and is then gone. The hub does keep a two-year record of AI counts and costs, which contains no words and is enforced to contain none, and you can switch on a conversation log that keeps the words for 90 days by default.

That is a decision worth making deliberately, so it has its own pages: What the hub keeps and What reMaestro Cloud keeps.