ProClaw ZeroWit

Automate your daily tasks hands-free through your AI companion, whether you are typing, talking, ProClaw ZeroWit proactively manages your projects so you can focus on real work.

Comment

VIDEO

https://youtu.be/dEgD8mRiZWY?si=Tx61UZVriC_RoSGZ

Project Name

ZeroWit

Elevator Pitch

Automate your daily tasks hands-free through your AI companion — whether you are typing, talking, or doing nothing at all, ZeroWit proactively manages your projects so you can focus on real work.

Inspiration

Two trillion dollars vanished from software stocks in early 2026. Atlassian down 35%. Salesforce down 40%. The per-seat SaaS model is dying because AI agents are replacing the humans who hold those seats.

Every SaaS company responded the same way: bolt AI onto the existing product. Nobody is building the layer that benefits from this shift.

We are two developers who spend two hours a day doing work about our work — tickets, time logs, status updates, sprint reports. The real SaaS moat was never the product. It was the habit. The muscle memory of learning where the buttons are.

What if the button didn't matter anymore?

What It Does

ZeroWit is a project management platform with a 3D AI avatar companion. You don't learn Jira. You don't learn Linear. You talk.

  • Zero-friction onboarding. Name and role. That is it. No admin setup. You are in.
  • Project dashboard. All projects in one view. Tool-agnostic — could sync from Jira, Linear, Asana, or a plain database. The user never knows what is underneath.
  • Smart kanban. Drag-and-drop with real-time sync, markdown, file attachments, mentions, priority badges.
  • AI semantic search. Natural language search across all projects using vector embeddings. Type what you mean, find what you need.
  • 3D AI companion. A VRM avatar with real-time lip sync driven by FFT audio analysis, eye tracking with micro-saccades, and facial expression blending. Not pre-baked. Every frame computed live.
  • Voice-first interaction. Talk to update your board. Status changes, task creation, and notifications from conversation.

The AI routes through OpenRouter to Qwen (multilingual) or GPT-4o (reasoning). Tools include Exa for web search, BrightData for scraping, Fal for image generation. ElevenLabs voices feed the avatar's lip sync. LiveKit and Agora handle real-time audio.

How We Built It

Next.js 16, React 19, TypeScript strict mode. Tailwind CSS + Framer Motion. Firebase Auth + Firestore for real-time data. The kanban board uses Firestore subscriptions — every card move propagates instantly.

The 3D system runs Three.js + pixiv VRM SDK + React Three Fiber. We built an LRU resource manager with reference counting to keep browser memory under control. Lip sync takes live audio, runs Web Audio API FFT, maps frequency bands to five visemes with exponential smoothing, updates blend shapes at 60fps.

OpenRouter routes all LLM calls. Gemini embeddings power semantic search against Firestore. Eight sponsor APIs connected through a unified pipeline — voice in, intelligence, tools, voice out, avatar animates.

Challenges We Ran Into

We cut features hard to ship end-to-end rather than half-build everything. Real-time lip sync without jitter required tuning exponential smoothing by hand. Firestore sync under rapid drag-and-drop caused race conditions we had to solve with careful ordering. VRM models eat memory — the LRU manager with a hard budget was necessary to prevent browser crashes.

Accomplishments We're Proud Of

The avatar is real. Production-grade 3D character with audio-driven lip sync, eye tracking, and expression blending. Runs client-side, no server GPU. Nobody else at this hackathon has this.

The kanban works. Not a mockup. Real-time drag-and-drop, markdown, attachments, mentions, semantic search. You can manage a real project with it today.

Eight sponsor technologies in one coherent product. Each one does something the others cannot.

What We Learned

People don't hate Jira because Jira is bad. They hate it because they have to learn it. Put AI between the human and the tool, the tool stops mattering.

A face changes everything. The same AI response feels different when a 3D character says it with eye contact and lip sync. People trust it more, engage longer, remember it.

Two people with the right AI tooling can ship what used to take a team of five a sprint.

What's Next

  • Proactive avatar: "You have three tasks in review and one blocker. Want me to ping the team?"
  • Tool-agnostic sync: plug in Jira, Linear, GitHub Issues. Switch backends, retrain nobody.
  • Custom avatars for enterprise. White-label the companion.
  • Per-action pricing, not per-seat. Revenue scales with value, not headcount.

Built With

agora, bright-data, elevenlabs, exa, fal, firebase, firestore, framer-motion, gemini, livekit, next.js, openai, openrouter, qwen, react, react-three-fiber, tailwind-css, three.js, typescript, vercel, vrm

Built With

Share this project:

Updates