⚠️ IMPORTANT: All features are experimental, under active development. Use at your own risk. Customization to your workflow required. © 2026 GLG, a.s. | ← Back to Index

6. Best Practices for AI Agents

6.1 When to Store

✅ Do store: - Decisions and their rationale - Facts learned during research - User preferences and corrections - Project context and relationships - Meeting outcomes and action items

❌ Don't store: - Temporary calculations or intermediate reasoning - Exact duplicates of previous entries - Raw data dumps (summarize first) - Sensitive data without user consent

6.2 How to Store Effectively

# BAD — too vague
uaml.learn("Did some work today")

# GOOD — specific, searchable, with context
uaml.learn(
    "Deployed v2.1 to production server 10.0.0.5. "
    "Migration took 12 minutes. No errors.",
    topic="deployment",
    source_type="observation",
    confidence=0.95
)

6.3 How to Recall Effectively

# BAD — too broad, wastes tokens
results = uaml.search("everything")

# GOOD — focused query with budget
context = uaml.recall(
    "What deployment issues have we had this month?",
    budget_tokens=800
)

6.4 Focus Engine Tuning

6.5 Token Budget Strategy

• Short answer needed: 200–500 tokens
• Detailed analysis: 500–1500 tokens
• Comprehensive report: 1500–3000 tokens
• Never exceed 50% of your context window for memory recall