Route smart-home assistant requests

01 · Role in the system

What Jev does here

Jev classifies the request category, target domain, device type, and requested device action in one speculative batch. It also checks whether the request contains multiple actions. Code ignores irrelevant speculative answers, splits compound requests through a separate language model, and sends conversational requests to that model instead of treating Jev as a text generator.

One speculative call replaces a routing cascade

Instead of chaining classifiers, the official demo batches the whole routing decision — request category, target domain, device type, requested action, plus whether the request contains multiple actions — into one speculative Jev call. Code then uses only the answers relevant to the request and ignores the rest.

The fallback paths are explicit. Compound requests are split by a separate language model, and conversational requests go straight to that model, because Jev is not being asked to write text. The result is a routing layer that costs one typed call for the common case and escalates only when the shape of the request demands it.

An architecture record, not a benchmark

The official page describes the architecture and the full source repository is promised at release, so this record currently verifies the described design rather than a deployed system. No accuracy, latency, device-coverage, or reliability numbers are published for the assistant.

Its value to the directory is as the reference pattern for request routing: parallel speculative judgments over a typed action taxonomy with code owning the branch selection. Community implementations can be compared against it, which is precisely what the surrounding home-domain records do.

02 · Control boundary

Where Jev sits

One request becomes parallel Choice and Noul judgments; deterministic code selects the applicable branch and owns device actions.

Code owns the loop, permissions, thresholds, validation, and side effects. Jev owns only the bounded judgments described above.

03 · Known limits

What this evidence does not prove

• The official page describes the demo architecture but says the full source repository will be available at release.
• The page publishes no accuracy, latency, device-coverage, or production-reliability benchmark for this assistant.

04 · Attribution

Public sources

This is an Official record: the example was published by TypeSafe itself.

• TypeSafe ↗Official · demo · public · checked 2026-09-18