QUICKSTART.md
QUICKSTART.md
Quick Start Guide - GAIA Experience Learning System
🚀 5-Minute Setup
1. Prerequisites Check
# Check Python (3.8+ required)
python --version
# Check pip
pip --version
2. Quick Install
# Clone if not already done
cd projects/week3/gaia-experience
# Install dependencies
pip install -r requirements.txt
# Setup AWorld (if needed)
cd AWorld && python setup.py install && cd ..
3. Configure API Keys
# Copy environment template
cp env.template .env
# Edit .env and add your OpenAI API key
# LLM_API_KEY=sk-your-key-here
4. Run Your First Demo
# Run the interactive demo
python demo.py
# Or use the shell script
./run.sh demo
📊 Quick Examples
Example 1: Index Existing Knowledge
# Index the GAIA validation dataset
./run.sh index
# This creates a searchable knowledge base from gaia-validation.jsonl
Example 2: Learn from Tasks
# Process first 5 tasks and learn from successes
./run.sh learn --start 0 --end 5
# Check learned experiences
ls experiences/
cat experiences/learned_experiences.json
Example 3: Apply Learned Knowledge
# Use learned experiences on new tasks
./run.sh apply --start 5 --end 10
# This will:
# - Load past experiences
# - Find relevant ones for each new task
# - Add them to the prompt for better performance
Example 4: Full Learning Loop
# Learn and apply simultaneously
./run.sh full --start 0 --end 20
# This enables both:
# - Learning from new successes
# - Applying past experiences
🎯 Common Use Cases
Use Case 1: Bootstrap with Validation Data
from knowledge_base import KnowledgeBase
kb = KnowledgeBase()
kb.index_gaia_validation('gaia-validation.jsonl')
# Now search for similar problems
results = kb.search("How to find papers on arXiv?", top_k=3)
for r in results:
print(f"Q: {r['question'][:100]}...")
print(f"A: {r['approach'][:100]}...")
Use Case 2: Custom Task with Learning
from experience_agent import ExperienceAgent
from AWorld.aworld.config.conf import AgentConfig, TaskConfig
from AWorld.aworld.core.task import Task
import asyncio
async def run_custom_task():
# Setup agent with learning
config = AgentConfig(
llm_provider="openai",
llm_model_name="gpt-5.6-luna",
llm_api_key="your-key"
)
agent = ExperienceAgent(
conf=config,
learning_mode=True,
apply_experience=True
)
# Create and run task
task = Task(
input="What is the capital of France?",
agent=agent,
conf=TaskConfig()
)
response = await agent.execute_task(task)
print(f"Answer: {response.answer}")
asyncio.run(run_custom_task())
📝 Quick Tips
Tip 1: Check Logs
# View latest logs
tail -f workspace/experience_agent_*.log
# Check for errors
grep ERROR workspace/*.log
Tip 2: Monitor Learning
# Watch experiences being added
watch -n 5 'wc -l experiences/learned_experiences.json'
Tip 3: Test Specific Task
# Run a specific task by ID
./run.sh test --task-id "c61d22de-5f6c-4958-a7f6-5e9707bd3466"
Tip 4: Reset and Start Fresh
# Clear all learned experiences and indices
rm -rf kb_index/ experiences/ workspace/
mkdir -p kb_index experiences workspace
# Start fresh
./run.sh full
⚡ Performance Tips
-
Batch Processing: Process multiple questions at once for efficiency
./run.sh full --start 0 --end 50 -
Preload Knowledge: Always preload for better initial performance
./run.sh apply --preload-kb -
Use Appropriate Models:
- Main agent:
gpt-5.6-solfor complex reasoning - Summarization:
gpt-5.6-lunafor cost efficiency
- Main agent:
-
Monitor Token Usage: Check logs for token consumption
grep "tokens" workspace/*.log
🔍 Troubleshooting
Issue: "API key not found"
Solution: Ensure .env file exists and contains valid LLM_API_KEY
Issue: "Module not found"
Solution: Install requirements: pip install -r requirements.txt
Issue: "AWorld not found"
Solution: Install AWorld: cd AWorld && python setup.py install
Issue: "Knowledge base empty"
Solution: Index validation data: ./run.sh index
📚 Next Steps
- Read the full README.md for detailed documentation
- Explore config.yaml for advanced configuration
- Check demo.py for programmatic examples
- Review the GAIA benchmark for context
🆘 Getting Help
- Check logs in
workspace/directory - Review error messages carefully
- Ensure all prerequisites are met
- Verify API keys are correct
Happy Learning! 🎉