QwenTree is a Tree File Agent built on Qwen Cloud's multimodal models. It organizes its capabilities as a navigable file system, where each skill is a file the agent can list, read, and execute.
🏆 Global AI Hackathon Series with Qwen Cloud 2026 — Track: Agent Society
qwentree> what do you see in this image?
│
▼
┌─────────────────────────────────────────────┐
│ 🧠 ORCHESTRATOR (Qwen3.7-Max) │
│ Analyzes → Routes → Invokes skills │
├──────────────────┬──────────────────────────┤
│ │ │
│ 🌳 TREE │ 🗄️ MEMORY (3 tiers) │
│ NAVIGATOR │ │
│ │ 🏛️ Mental Models (Chroma)│
│ skills/vision/ │ 📝 Observations (PgSQL) │
│ skills/code/ │ 📦 Raw Facts (Redis) │
│ skills/web/ │ │
│ skills/audio/ │ 🧠 LIVING BRAIN │
│ skills/video/ │ (real sessions ingested) │
│ skills/files/ │ │
│ ... │ │
└──────────────────┴──────────────────────────┘
| Feature | How QwenTree does it |
|---|---|
| 🏛️ Tree-File Architecture | Skills as navigable files — the agent ls, cd, cat over its own capabilities |
| 👁️ True Multimodal | Not just text — image, audio, video, code, web — fully integrated |
| 🧠 Living Memory | 3-tier memory (ChromaDB + PostgreSQL + Redis) populated with 6,099 items from 212 real development sessions |
| 🔧 Extensible like SHIVA | Add a skill = create a .py file in the right folder |
| 💻 CLI-first with Textual TUI | Native terminal with rich formatting, panels, autocomplete |
| 🔗 Ecosystem Integration | Connects with n8n, SHIVA, Academy, databases |
qwentree/skills/
├── vision/ 👁️ Analyze images, OCR, URL analysis (Qwen-VL)
├── audio/ 🎤 Transcribe speech, synthesize voice (Qwen-Audio, CosyVoice)
├── video/ 🎬 Analyze videos, extract frames (Qwen-VL + FFmpeg)
├── code/ 💻 Execute Python, analyze code, refactor
├── web/ 🌐 Search web, fetch pages (Playwright)
├── files/ 📁 Read, write, search, tree, find by content
├── system/ ⚙️ System info, shell commands
├── memory/ 🧠 Recall past sessions, search 3-tier memory
├── media/ 🎨 Generate images (Wan) & videos (HappyHorse #1)
└── integrations/ 🔗 SHIVA, n8n, Academy
QwenTree uses a 3-tier hierarchical memory system inspired by cognitive science:
| Tier | Storage | Content | Size |
|---|---|---|---|
| 🏛️ Mental Models | ChromaDB | Canonical facts extracted from sessions | 3,046 items |
| 📝 Observations | PostgreSQL | Patterns, decisions, learnings | 2,751 items |
| 📦 Raw Facts | Redis | Full session transcripts with metadata | 212 sessions |
| TOTAL | 6,099 items |
The memory was auto-populated from 212 real development sessions, making the agent functional from day one.
# 1. Clone and setup
git clone https://github.com/SebaWag/qwentree.git
cd qwentree
# 2. Environment
cp .env.example .env
# Edit .env with your QwenCloud API key
# 3. Dependencies
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
playwright install chromium
# 4. Infrastructure (Docker)
docker compose up -d # Redis + PostgreSQL + ChromaDB
# 5. Populate memory (ingest sessions)
python3 -m scripts.ingest_brain
# 6. Run!
python3 -m qwentree# Launch TUI
python3 -m qwentree
# Commands inside TUI
/help # Show all commands
/ls # List all skill categories
/tree # Show full skill tree
/stats # Show system statistics
/mode qwen|local # Switch API mode
# Or use direct CLI mode
python3 -m qwentree --cli| Component | Technology |
|---|---|
| 🧠 LLM Orchestrator | Qwen3.7-Max / Qwen3.7-Plus |
| 👁️ Vision | Qwen-VL-Max |
| 🎤 Audio | Qwen-Audio + CosyVoice-01 |
| 🎬 Video Generation | HappyHorse-1.1-T2V (#1 worldwide) |
| 🖼️ Image Generation | Wan |
| 🧠 Memory | ChromaDB + PostgreSQL + Redis |
| 🖥️ CLI | Textual (Python TUI framework) |
| 🌐 Web | Playwright + httpx + BeautifulSoup |
| 🐳 Infra | Docker Compose |
📁 Files: 48 Python files
📝 Lines: 3,818 lines of code
🧩 Skills: 31 skills in 10 categories
🧠 Memory: 6,099 items (3 tiers)
🧪 Tests: 89 tests (all passing)
Apache 2.0 — See LICENSE for details.
Built with ❤️ for the Global AI Hackathon Series with Qwen Cloud 2026
🌳 QwenTree on GitHub