Design the agent runtime for Platform Mastery AI Agents — the system that turns ingested knowledge into intelligent responses:

Components:
1. **Query understanding**: Classify user intent (question, action request, code help, debugging)
2. **RAG retrieval**: Multi-stage retrieval (keyword → semantic → reranker)
3. **Context assembly**: Build the optimal prompt with retrieved knowledge
4. **Reasoning engine**: Claude/Gemini for generating response
5. **Tool execution**: MCP tools for taking platform actions
6. **Response formatting**: Code blocks, links, step-by-step guides
7. **Memory**: Remember user context across conversations
8. **Feedback loop**: Learn from user corrections

For each component: architecture, model selection, cost, latency target.
Total query latency target: <3 seconds for simple questions, <10 seconds for complex actions.