Overview & result
A custom conversation integration for Home Assistant Assist that uses TypeSafe's System One API (the Jev model family) to classify spoken or typed requests, resolve entities and control devices without generating text. The README describes sub-200ms decision routing and a sibling project, home-assistant-laya, that runs a local open-weight model instead.
From the original post
Home Assistant conversation integration powered by the Jev / TypeSafe AI API for fast, structured intent routing and device control
How Jev fits in the loop
- An utterance from Home Assistant Assist is tokenised, and numeric slots such as brightness or temperature are extracted.
- Candidate devices and intent handlers are retrieved with BM25 ranking or exhaustive evaluation.
- TypeSafe's System One API classifies the intent and binds the target entities without generating text.
- Home Assistant executes the resulting device action.
How to reproduce
- Inspect the original showcase and source material at https://github.com/allenporter/home-assistant-typesafe
- Verify the bounded prompt and input candidate schema configured for Jev
- Benchmark decision latency and classification accuracy against baseline models
Why this build matters
It is the cloud counterpart to the same author's home-assistant-laya, which runs an open-weight model locally, so the two can be compared on the same pipeline.
Limitations
- It is a cloud integration: requests leave the home network for the TypeSafe API.
- The sub-200ms figure is the author's description, with no benchmark in the README opening.
- It is a custom integration, not part of Home Assistant core.