igny8/agent58k/QUICKSTART.md

Quick Start Guide

Installation (30 seconds)

cd ~/agent58k
chmod +x setup.sh
./setup.sh

Start Server

cd ~/agent58k
source venv/bin/activate
python server.py

Configure IDE

Continue.dev

Copy the config:

cp continue-config.json ~/.continue/config.json

Cursor

Settings → Models → Add Custom:

• Base URL: http://localhost:8000/v1
• Model: zdolny/qwen3-coder58k-tools:latest

Test It Works

chmod +x test_server.sh
./test_server.sh

Example Prompts

Instead of: ❌ "Can you help me read the config file?"

Use: ✅ "Read config.py and show me the database settings" ✅ "Search the codebase for all files importing requests" ✅ "Create a new file utils/parser.py with a JSON parser" ✅ "Run the tests in tests/ directory" ✅ "Find all TODO comments in Python files"

What Makes This Different

Your original code had:

• ❌ Only 2 basic tools (calculate, python)
• ❌ Broken tool extraction (regex parsing)
• ❌ No streaming support
• ❌ No file operations
• ❌ No terminal access
• ❌ Single-step execution only

This version has:

• ✅ 8 comprehensive tools (file ops, terminal, search, etc.)
• ✅ Proper Qwen Agent tool integration
• ✅ Full streaming support
• ✅ Multi-step agent reasoning
• ✅ Works like native Cursor/Continue
• ✅ Production-ready error handling

Auto-Start on Boot (Optional)

# Edit the service file and replace %YOUR_USERNAME% and %HOME%
sudo cp qwen-agent.service /etc/systemd/system/
sudo systemctl enable qwen-agent
sudo systemctl start qwen-agent

Troubleshooting

"ModuleNotFoundError: No module named 'qwen_agent'" → Activate venv: source venv/bin/activate

"Connection refused to localhost:8000" → Start server: python server.py

"Ollama API error" → Start Ollama: ollama serve → Pull model: ollama pull zdolny/qwen3-coder58k-tools:latest

Agent not using tools → Be explicit: "Use the file read tool to..." → Check server logs for errors

What Fixed

1. Tool System: Implemented proper BaseTool classes that Qwen Agent understands
2. Streaming: Added SSE support with proper chunk formatting
3. Response Handling: Properly extracts content from agent responses
4. Multi-step: Agent can now chain multiple tool calls
5. Error Handling: Comprehensive try/catch with detailed error messages
6. IDE Integration: OpenAI-compatible API that works with Continue/Cursor

Files Created

• server.py - Main server (400+ lines with 8 tools)
• requirements.txt - Python dependencies
• setup.sh - One-command installation
• test_server.sh - Verify everything works
• continue-config.json - IDE configuration
• qwen-agent.service - Systemd service
• README.md - Full documentation
• QUICKSTART.md - This file

Next Steps

1. Run setup: ./setup.sh
2. Start server: python server.py
3. Configure your IDE (copy continue-config.json)
4. Test with: "List files in the current directory"
5. Try complex tasks: "Read all Python files, find bugs, fix them"

Enjoy your fully-capable AI coding assistant! 🚀