Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

Yappy

macOS voice agent where Jev picks each step's operation and target control, escalating to an LLM agent when unsure.

Articleproduction
Open source ↗

This summary is based on how curated lists describe the project; its README was not available to review.

Overview

Yappy is a closed-source macOS voice agent with a public write-up including measurements. According to the list summaries, on each step it asks Jev one Choice covering the operation and the target control from the front window's accessibility table. Only validated, high-confidence answers are executed; a chat model writes text only when typing is needed; and a full LLM agent takes over on low confidence, actions with no effect, or unknown field values.

The author reports 275 to 690 ms per decision and $0.003 for five decisions on the hosted plan. It is a concrete example of putting a decision model at the center of computer use with an LLM fallback.

What the decision model decides

TypeQuestionAnswers
ChoiceWhich operation and target control should be used next?—

How it works

  1. Read the front window's accessibility table.
  2. Jev chooses operation and target.
  3. Execute only validated, high-confidence answers.
  4. Escalate to a full LLM agent on low confidence or no effect.

Strengths

  • Author-reported 275 to 690 ms per decision.
  • Low cost per step.
  • Clear escalation rules.

Caveats

  • Closed source.
  • Measurements are author-reported.

Getting started

See https://yappy.biz/jev

At a glance

Type
Article
Status
production
Runtime
macOS app

Use cases and integrations

How curated lists describe it

  • Computer use: macOS voice agent that asks Jev one `Choice` per step (operation and target control) over the front window's accessibility table, executes only validated high-confidence answers, and escalates to a full LLM agent on low confidence, no-effect actions, or unknown field values; author-reported 275–690 ms per decision.

    yibie/awesome-jev

  • macOS voice agent (closed source, public write-up with measurements). On its hosted plan Jev picks the operation and target control from the window's accessibility table each step; a chat model writes text only for typing, and the full agent takes over when confidence drops. Author-reported: 275–690 ms per decision, $0.003 for five.

    AnotiaWang/awesome-decision-models

Frequently asked questions

What does Jev decide?

The operation and target control for each step.

When does the LLM take over?

On low confidence, no-effect actions or unknown field values.

Is it open source?

No, but there is a public write-up.

Related projects

  • jev-voice-control★ 6

    macOS menu-bar Swift app that turns on-device transcribed speech into Jev typed decisions and macOS actions, confirming hard-to-undo steps.

    AppbetaSwift
  • jev-browser-pilot

    Python decision layer for browser and desktop automation: code observes, acts and verifies; the model makes one typed choice per step.

    ProjectbetaPython
  • jev-mobile

    Durable Android sub-agent: a SQLite-backed worker observes the UI and Jev picks among valid actions, with verification and escalation.

    MCP serverbetaPython
  • kureksistant

    Voice assistant daemon for Hyprland Linux desktops that acts through tool modules, with Jev gating which requests and actions proceed.

    ProjectbetaPython
  • computer_use

    Go macOS controller where Jev picks the next accessibility action, the target element and whether the goal is done.

    ProjectprototypeGo
  • Jev Voice

    Talk to your Mac: local whisper.cpp transcribes, one Jev request picks the action and typed arguments, and code executes macOS automation.

    ProjectprototypePython