Ovo

Open-source proactive AI desktop assistant.
Watches your screen, suggests next steps, runs 100% on your machine.

English · 简体中文

⬇ Download for macOS · 📖 Philosophy · 💬 Discussions · 🤝 Contributing

Ovo watching a Gmail draft, predicting the reply, and offering to copy it — all visible, all auditable.

🪟 What is Ovo?

Ovo is a proactive AI desktop assistant that watches what you're doing — and helps before you have to ask.

Most AI tools today wait for you to type a prompt. Ovo flips that:

• It observes your screen every few seconds (OCR + window context).
• It understands your intent through a multi-pass reasoning pipeline with a long-term knowledge graph.
• It suggests the next step proactively — drafting an email reply, copying a snippet, reminding you of a deadline.
• It shows you every step of its thinking in a glass-box timeline — no magic, no black box.
• And it learns from your feedback — accept, reject, or teach Ovo to never do that again.

Built for people who want an AI that's actually present in their day — without giving up control, transparency, or privacy.

✨ Why Ovo? — The 5 things that make Ovo different

🚀 Proactive, not reactive

While ChatGPT waits for you to type, Ovo notices you're drafting an email to a customer and silently prepares the reply before you ask. You see the suggestion, decide whether to use it. Zero prompting overhead.

🪟 Glass-box transparency

Every other AI is a black box. Ovo shows you:

• 📸 What it saw (the OCR text from your screen)
• 🧠 What it thought (the full prompt sent to the LLM)
• 💡 What it decided (the structured suggestion + confidence)
• ⚡ What it did (every action with output and duration)

Click any suggestion → see its full reasoning chain. No more "why did the AI say that?"

🎓 Teachable, not opinionated

Don't like a suggestion? You can:

• Reject this one — Ovo remembers and won't repeat the mistake
• "Never do this again" — Ovo writes the rule into its knowledge graph
• Adjust trust level — give Ovo more or less autonomy per action type

Ovo gets smarter from your feedback, not from cloud telemetry.

🔒 Local-first, BYO LLM

• Screenshots and OCR happen on your machine — they never leave
• Bring your own Claude / OpenAI / local LLM key — no proxy server
• Built-in sensitive data redaction (API keys, JWT, credit cards, ID numbers)
• App-level blacklist (1Password, banking apps, etc. — never observed)
• Hard pause (5/15/60 min) when you need privacy

🧠 Long memory through a knowledge graph

Ovo builds a personal knowledge graph as you work:

• Entities (people, projects, documents, concepts)
• Relationships between them
• Memory events with timeline + intent tagging
• Personality profile that evolves with your patterns

Open the Memory panel to see exactly what Ovo knows about you. Edit it. Pin important entities. Delete anything you don't want remembered.

🆚 Ovo vs. Other AI Tools

📥 Install

Option 1: Download (recommended)

First launch on macOS: until we ship a signed/notarized build, you'll see "Ovo can't be opened because the developer cannot be verified". Right-click the app → Open → Open. Only needed once.

Windows / Linux: planned for v0.5+. Star the repo to get notified.

Option 2: Build from source

git clone https://github.com/dushaobindoudou/ovo.git
cd ovo
pnpm install
pnpm dev          # Dev mode (Vite + Electron live-reload)
# or
pnpm pack:mac     # Build production DMG to out/

Requirements: Node 20+, pnpm 10+, macOS (Apple Silicon or Intel)

First-run setup

1. Grant screen recording permission — Ovo will guide you to System Settings
2. Configure your AI backend — choose Claude Code / OpenClaw / Hermes / direct API
3. Set your API key (encrypted with macOS Keychain, never sent anywhere)
4. Tell Ovo about you — optional 4-step onboarding to seed the knowledge graph

That's it. Within 1-2 minutes Ovo will start observing and suggesting.

🖼️ Screenshots

Screenshots and GIFs will be added in v0.2 release. PRs welcome — see #good-first-issue.

🏗 Architecture

┌─────────────────────────────────────────────────────────────┐
│  Renderer (React + Zustand)                                 │
│  ┌──────────────┐  ┌────────────────┐  ┌────────────────┐  │
│  │ Main Console │  │ Floating Icon  │  │ Suggestion     │  │
│  │ (#console)   │  │ (#float)       │  │ Panel/Toast    │  │
│  └──────┬───────┘  └───────┬────────┘  └───────┬────────┘  │
└─────────┼──────────────────┼───────────────────┼───────────┘
          │   IPC (preload.cjs, context-isolated)
┌─────────▼──────────────────▼───────────────────▼───────────┐
│  Electron Main Process                                      │
│  ┌──────────────┐  ┌────────────────┐  ┌────────────────┐  │
│  │ AutoCapture  │→ │ OCR Engine     │→ │ Event Processor│  │
│  │ (5s)         │  │ (Tesseract)    │  │ (intent infer) │  │
│  └──────────────┘  └────────────────┘  └───────┬────────┘  │
│                                                ▼            │
│  ┌──────────────────────────────────────────────────────┐  │
│  │ Multi-Pass Prompt Engine                             │  │
│  │ Pass 1: Observe (intent + entities)                  │  │
│  │ Pass 2: Synthesize (predictions + actions)           │  │
│  └────────────┬─────────────────────────────────────────┘  │
│               ▼                                              │
│  ┌──────────────────────────────────────────────────────┐  │
│  │ Agent Bridge (Claude Code / OpenClaw / Hermes / API) │  │
│  └────────────┬─────────────────────────────────────────┘  │
│               ▼                                              │
│  ┌──────────────┐  ┌────────────────┐  ┌────────────────┐  │
│  │ Knowledge    │  │ Action         │  │ Feedback       │  │
│  │ Graph (SQL)  │  │ Executor       │  │ Engine         │  │
│  └──────────────┘  └────────────────┘  └────────────────┘  │
└─────────────────────────────────────────────────────────────┘

Tech stack: Electron 34 · React 19 · TypeScript · Vite · Tailwind · better-sqlite3 · Tesseract.js · Zustand

See docs/PRODUCT_PHILOSOPHY.md for the design philosophy and docs/ELECTRON_IPC_MAPPING.md for IPC contracts.

🗺 Roadmap

Shipped ✅

• Multi-pass prompt engine (observe → synthesize)
• 4 AI backends (Claude Code / OpenClaw / Hermes / direct API)
• Knowledge graph with entities / relationships / memory events
• Pipeline transparency (every step logged + queryable)
• Privacy controls (pause, blacklist, sensitive data redaction)
• Prompt self-eval (Ovo reflects on its own pipeline quality)
• Bootstrap onboarding (4-step seed)

In Progress 🚧

• Trust ladder UI (per-action autonomy levels: show → draft → confirm → auto + 5s undo → full delegation)
• Glass butler floating window (real-time "Ovo is doing X because Y" with [Let it] [Don't] [Never again] buttons)
• AI behavior timeline as the main view (currently buried in a second-level panel)

Planned 🔮

• Windows + Linux support
• Plugin/extension system
• KPI dashboard (TTFV, hit rate, undo rate — see PRODUCT_PHILOSOPHY.md)
• macOS Notarization + auto-updates
• Multi-language UI (currently 中文/English mixed)

See open milestones for what's coming next.

📚 Documentation

Start here

Building / using

Project state

💬 Community

We're building Ovo in the open. Come help us shape it:

• 🐙 GitHub Discussions — questions, ideas, show & tell
• 🐛 Issue tracker — bugs and feature requests
• 🐦 Twitter @dushaobin — build-in-public updates
• 💚 WeChat group (中文社区) — see README_CN.md for QR code

🤝 Contributing

Contributions are warmly welcomed — from typo fixes to new features.

• First-time contributor? Look for good first issue — each one comes with full context, file pointers, and acceptance criteria.
• Got a bigger idea? Open a Discussion first so we can align on direction.
• Pull request? See CONTRIBUTING.md for dev setup, code style, and PR workflow.

Every external PR gets a review within 24 hours and a Twitter thank-you. 🙏

🔐 Security

Found a vulnerability? Please do not open a public issue. See SECURITY.md for our responsible disclosure process.

💖 Support

If Ovo saves you time, consider supporting development:

• ⭐ Star this repo — the cheapest way to help
• 🐦 Share on Twitter
• ☕ Buy me a coffee (international) or 爱发电 (China)
• 🤝 Contribute code, docs, translations, or designs

📜 License

MIT © 2026 dushaobin

Ovo is and will always be open source and free for personal use. Commercial fork friendly under MIT.

🙏 Acknowledgements

Ovo stands on the shoulders of giants:

• Electron — desktop runtime
• React — UI framework
• Anthropic Claude — default AI backend
• Tesseract.js — OCR engine
• better-sqlite3 — knowledge graph storage
• Lucide — icon library
• Tailwind CSS — styling
• And every contributor whose name will appear here ❤️

⭐ Star History