Load preset
Context allocation
0% used
⚡ Context Engineering Tips
- Keep your system prompt under 10% of total context for most use cases
- RAG chunks should be trimmed to only the most relevant passages — don't dump all results
- Reserve at least 15–20% for model output, more for code generation
- Tool schemas can eat 2–5K tokens each — audit regularly
- Chat history grows silently — implement sliding window or summarization