abstractassistant 0.4.1

A sleek (macOS) system tray application providing instant access to LLMs

AbstractAssistant

One-click, tray-accessible agent host for AbstractFramework.

AbstractAssistant runs agentic loops (ReAct/CodeAct/MemAct) on top of:

• AbstractAgent (agent patterns)
• AbstractRuntime (durable runs, waits, ledgers)
• AbstractCore (providers + tools + media handling)
• AbstractVoice (STT/TTS)

Docs:

• docs/getting-started.md
• docs/architecture.md

Install

pip install "abstractassistant"

Quick start

Tray (macOS):

assistant tray

Alias:

abstractassistant tray

Terminal (one turn):

assistant run --prompt "What is in this repo and where do I start?"

Provider/model override:

assistant run --provider ollama --model qwen3:4b-instruct --prompt "Summarize my changes"

Tool approvals (important)

AbstractAssistant enforces a durable tool boundary:

• read-only / known-safe tools can auto-run
• anything else pauses and requires approval (tray dialog or terminal prompt)

This aligns with the framework’s durability + safety model: tools are executed by the host, not persisted as callables inside run state.

Data & durability

By default, assistant state is stored in ~/.abstractassistant/ (configurable via --data-dir):

• session.json: fast UI snapshot (transcript + last run id)
• runtime/: run store + ledger + artifacts (source of truth)

Development

pip install -e ".[dev]"
python -m pytest -q
assistant tray --debug

• 📱 Unobtrusive: Lives quietly in your menu bar until needed
• 🔊 Conversational: Optional voice mode for natural AI interactions

📚 Documentation

Guide	Description
📖 Installation Guide	Complete setup instructions, prerequisites, and troubleshooting
🎯 Getting Started Guide	Step-by-step usage guide with all features explained
🏗️ Architecture Guide	Technical documentation and development information

📋 Requirements

• macOS: 10.14+ (Mojave or later)
• Python: 3.10+
• Qt Framework: PyQt5, PySide2, or PyQt6 (automatically detected)
• Core deps: AbstractAgent + AbstractRuntime + AbstractCore + AbstractVoice (installed with abstractassistant)
• Audio note: audio attachments are auto-transcribed via AbstractVoice (first run may download model weights)
• Video note: frame-sampling fallback may require ffmpeg on your PATH

🤝 Contributing

Contributions welcome! Please read the architecture documentation and follow the established patterns:

• Clean Code: Follow PEP 8 and use type hints
• Modular Design: Keep components focused and reusable
• Modern UI/UX: Maintain the sleek, native feel
• Error Handling: Always include graceful fallbacks
• Documentation: Update docs for any new features

📄 License

MIT License - see LICENSE file for details.

🙏 Acknowledgments

AbstractAssistant is built on excellent open-source projects:

Core Dependencies

• AbstractCore: Universal LLM interface - enables seamless multi-provider support
• AbstractVoice: High-quality text-to-speech engine with natural voice synthesis

Framework & UI

• PyQt5/PySide2/PyQt6: Cross-platform GUI framework for the modern interface
• pystray: Cross-platform system tray integration
• Pillow: Image processing for dynamic icon generation

Part of the AbstractX Ecosystem

AbstractAssistant integrates seamlessly with other AbstractX projects:

• 🧠 AbstractCore: Universal LLM provider interface
• 🗣️ AbstractVoice: Advanced text-to-speech capabilities

See ACKNOWLEDGMENTS.md for complete attribution.

---

Built with ❤️ for macOS users who want AI at their fingertips