AssistantXAssistantX

The road to the
assistant that acts.

Every phase ships real capability. No vaporware — if it's not on this page, it's not being built yet.

v0.5Shipped

The Foundation

Voice-first MVP

  • Wake-word voice activation — noise-adaptive threshold, second-opinion check on quiet speech, and voice matching against your own profile
  • Natural conversation — routing across local and cloud models, with follow-up listening after each reply
  • Speech that streams — replies are spoken sentence by sentence instead of after the full answer
  • Conversation memory — every exchange stored locally, with rolling summaries for older context
  • Workspace — projects, blueprints, missions, memory, lessons, skills and workflows, each with its own view
  • Desktop app for Windows with a guided setup wizard
v0.6In progress

The Operator

From answering to doing

  • A small always-resident model decides what you meant and acts on it, without waking the large one
  • 41 actions across 16 tool groups — apps, web, media, volume, files, clipboard, screenshots, reminders, power
  • Risky actions ask first — shutting down, closing an app or typing on your behalf needs a spoken yes
  • Long jobs hand off to the larger model in the background, and report back when finished
  • Task registry and a planning step ahead of execution — in progress
v1.0Up next

The Intelligence Layer

Reasoning and judgement

  • A review pipeline that checks every action before it runs, not only the risky ones
  • Simulation Engine, Reality Check, Trust Engine and Devil's Advocate — written, waiting to be wired into the decision path
  • Confidence scoring surfaced on every answer, not left blank
  • Skill Evolution — abilities version, improve and roll back like code
  • Failure & Success Analysis — AssistantX learns from every outcome
  • AI Constitution — safety rules injected into every decision
v2.0Planned

The Orchestrator

Multi-agent orchestration

  • Multi-Agent System — specialist agents working in parallel
  • Deep Research — autonomous multi-source investigation
  • World Model — predictive understanding of your digital environment
  • Knowledge-gap detection — AssistantX knows what it doesn't know
  • Goal decomposition — long-horizon missions broken into verified steps
  • Cross-device intelligence — one brain, every screen

Want to shape what's next?