TypeSafe Conversation (Jev) for Home Assistant

A conversation agent for Home Assistant Assist that uses TypeSafe's Jev to interpret commands.

Independent community project. This integration is not affiliated with, endorsed by, or supported by TypeSafe AI. It is an unofficial, community-maintained client for their public API, and you use it with your own TypeSafe account and API key. "TypeSafe", "Jev" and the TypeSafe logo are trademarks of TypeSafe AI and are used here for identification purposes only. Questions or issues about this integration belong in this repository's issue tracker, not with TypeSafe.

Jev is a System One model: it does not generate text. It answers typed questions about a piece of state and returns choices with calibrated probabilities. This integration sends each Assist command, together with the entities exposed to Assist, to Jev in a single request containing a speculative fan-out of questions (intent, target, device, room, brightness, and so on). The answers are mapped onto Home Assistant's built-in intents (HassTurnOn, HassLightSet, HassGetState, ...), which do the actual device control. Spoken confirmations come from templates.

When Jev is unsure, the request is conversational, or it contains several actions, the original text is handed to an optional fallback conversation agent (for example an OpenAI, Anthropic or Ollama agent).

Ambiguous device references ("turn off the fan" when there are two fans) are resolved without the fallback: unavailable devices are skipped, the speaker's room is preferred when the command comes from a satellite, plurals target all devices of that kind, and otherwise the agent asks "Which fan: A or B?" and resolves the reply with a second, tiny Jev request.

Installation

HACS (Recommended)

1. Make sure HACS is installed.
2. Click the badge above, or go to HACS → Integrations → ⋮ → Custom repositories and add:

   https://github.com/JanOstrowka/typesafe-assist
   

   with category Integration.
3. Download "TypeSafe Conversation (Jev)" and restart Home Assistant.

Manual

Copy custom_components/typesafe_conversation into your config/custom_components/ directory and restart Home Assistant.

Setup

1. Settings → Devices & services → Add integration → TypeSafe Conversation (Jev).
2. Enter your TypeSafe API key.
3. Open the entry's options and set a fallback conversation agent (recommended), plus confidence thresholds if needed.
4. Settings → Voice assistants → pick or create a pipeline and set its conversation agent to the new Jev entity. Keep "Prefer handling commands locally" enabled so exact sentence matches stay on-device.

Every request is visible in the Assist debug view with Jev's latency, top intent probabilities, confidence, and the slots sent to Home Assistant.

Notes

• Only entities exposed to Assist are sent to TypeSafe.
• Spoken responses are English.
• Intents that need free text (shopping list, broadcast) are routed to the fallback agent, since Jev cannot extract free text.
• Requires Home Assistant 2025.6 or newer.

Disclaimer

This is an independent open-source project and is not affiliated with TypeSafe AI. Use of the TypeSafe API is subject to TypeSafe's own terms and pricing. All product names, logos and brands are property of their respective owners.