How does this differ from the local sibling integration?

TypeSafe handles fast structured routing with no local GPU or CPU load, while the sibling home-assistant-laya runs open-weight models on local CPU, CUDA, or Apple Silicon using the same five-stage pipeline shape.

What does the five-stage pipeline actually do?

Tokenization, candidate retrieval, hydration, engine scoring, and target binding run in order. Retrieval prunes with lexical or exhaustive search, then one Jev round trip scores intent, target, and compound questions.

When does a voice command fall back instead of executing?

Below the confidence threshold or above the compound threshold it escalates to the configured System 2 agent, preserving domain and area context so identically named devices do not collide.

What setup and tuning options should I expect?

Set the API key and model defaulting to jev-latest during setup, assign TypeSafe as the Assist conversation agent, then tune confidence, compound, retriever, domain filter, and fallback options.

How do I configure the integration?

Install the custom component into custom_components/typesafe and add it via Settings > Devices & Services > Add Integration > TypeSafe, supplying an API key and optionally a model (default jev-latest). Then set the Assist pipeline's Conversation agent to TypeSafe.

What options can I tune?

confidence_threshold (default 0.40, below which it escalates), compound_threshold (default 0.70 for multi-action utterances), retriever_type (lexical BM25 or exhaustive), domain_filter_mode (none, strict, or boost), and an optional fallback_agent.

What goes to the cloud?

Every utterance is sent to the TypeSafe System One API for scoring, which is why it runs cloud-accelerated with zero local GPU or CPU overhead. The sibling home-assistant-laya is the 100% local, offline-capable alternative.

CommunityJev Agents & MCP

Home Assistant TypeSafe

Home Assistant conversation integration that routes Assist utterances with Jev Choice and Noul scoring, binding intents and entities locally before falling back on harder requests.

See highlights

Open github.com

Stars: 2
Language: Python
License: APACHE-2.0
Published by: Community
Author: allenporter
Last verified: Oct 2026

allenporter/home-assistant-typesafe

User repository on GitHub

View on GitHub

Home Assistant conversation integration powered by the Jev / TypeSafe AI API for fast, structured intent routing and device control

Forks: 2
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

A Home Assistant custom conversation integration powered by TypeSafe AI System One (Jev model family) for fast, structured intent routing, entity resolution, and device control. Developed alongside its sibling project home-assistant-laya (a 100% local, open-weight implementation running directly on local hardware), this integration provides cloud-accelerated System One decision routing with zero local GPU/CPU overhead.

Quoted from the project's README. Read the full README.

Tagscommunity github

Works forJev for Routing

Used inJev for Software & DevTools

Languages: Python

Highlights

Routes natural-language commands against registered intent schemas and exposed entities using Choice and Noul primitives.
Retrieves candidates lexically or exhaustively with domain filtering to scale across homes with many entities.
Binds entities or areas with confidence gating, domain matching, continuous numeric slots, and area-aware disambiguation.
Escalates low-confidence and multi-action utterances to a fallback agent instead of executing directly.

Watch out

Cloud-hosted routing needs a TypeSafe key and network access; low-confidence or compound requests still depend on a configured fallback agent. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

agent-harness-with-jev-llm

Demo of the LangChain Jev harness pattern in LangGraph: a fast classifier routes events and gates a risky tool call, leaving open-ended reasoning to a chat LLM. Entry is POST /events; output lands in JSONL files.

Jev Agents & MCP#community#github

Community

dhirajpatra/agent-harness-with-jev-llmPython

jeveloper

Claude Code plugin that puts Jev judgments in the agent loop to route tool calls, verify outputs, judge completion, and pick next actions with metered calls.

Jev Agents & MCP#community#github

Community

pur4v/jeveloperPython

langgraph-jev-router

LangGraph conditional edge that uses Jev parallel Choice routing between specialized agents instead of a generative model.

Jev Agents & MCP#community#github

Community

JTankariya/langgraph-jev-routerPython

vexjoy-agent

441

A four-layer agent toolkit for Claude Code, Codex, Factory, and Reasonix: 43 agents, 62 skills, 78 hooks, 147 scripts, and a Jev-backed router that gates dispatch on intent alignment.

Jev Agents & MCP#community#jevbest

Community

notque/vexjoy-agentPython

Jevify

An agent skill from the ThursdAI community that teaches Choice, Noul, and Score batching, scans repos for movable LLM judgments, and ships Python helpers without extra npm deps.

Jev Agents & MCP#community#agents#cookbooks

Community