Your Personal AI Assistant with Bounded Autonomy

A personal agentic AI that combines planning, memory, and tool connections to prepare tasks independently and request your approval before consequential actions.

$ python -m agent run "calculate (12 + 8) * 3"
🧠 Planning task: calculate (12 + 8) * 3
✅ Calculated: 60
Task completed successfully
$ python -m agent run "remember Learn Python for 20 minutes"
💾 Memory saved: "Learn Python for 20 minutes"
$ python -m agent web
🌐 Opening liquid-glass interface at http://127.0.0.1:8765

What It Does

🧠 Smart Planning

Deterministic offline planner with strict JSON model-plan validation. Breaks down your goals into safe, manageable steps without asking permission for routine actions.

💾 Persistent Memory

Local SQLite notes with Unicode-aware search. Save and retrieve information across sessions with encrypted, private storage on your device.

🔧 Tool Connections

Explicit allowlisted tools for calculations, searches, and local operations. Up to 4 validated steps per task with safe, reversible preparation.

🛡️ Risk-Based Permissions

Requests approval only for consequential actions like messages, payments, or deletions. Final drafts are presented for explicit review before execution.

🌐 Flexible Operation

Works both online and offline. Choose between local models or cloud providers with explicit configuration - no silent fallbacks or unexpected connections.

👁️ Transparent Review

See exactly what your AI plans to do before it happens. Edit, reject, or approve actions with single-use tokens that invalidate on changes.

How It Works

1

Set Your Goal

Give your AI a clear objective. It will break it down into specific, actionable steps while respecting your configured capabilities and boundaries.

2

Safe Preparation

The AI automatically performs safe, reversible tasks like calculations, searches, and local memory access without asking for permission each time.

3

Final Review

Before any consequential action, you receive an exact proposal. Review, edit, or reject drafts with full transparency and control.

4

Approve & Execute

Once approved, the action is executed. Single-use approvals ensure you maintain control over exactly what happens and when.

Who It's For

👨‍💼

Busy Professionals

Delegate routine tasks and information management while maintaining full control over important decisions and communications.

🧑‍🎓

Students & Researchers

Organize study materials, manage research notes, and automate calculations while keeping your data private and secure.

👨‍💻

Developers & Tech Users

Enhance productivity with a local-first AI that respects your privacy and provides transparent, predictable assistance.

🏢

Privacy-Conscious Users

Maintain complete control over your data with offline operation and optional local models, no cloud processing required.

Ready to Try Personal-agentic-ai?

Experience the power of bounded autonomy with your personal AI assistant. Start with safe, controlled automation today.

Open the App