Quick Reference
Use this page when you need fast recall during implementation or review sessions.
It condenses the key components, signals, and decision rules from the learning path.
Core symbols and functions
| Symbol or function | Location | Purpose |
|---|---|---|
chat |
src/api/server.py |
Public endpoint for prompt processing |
handle_prompt |
src/core/engine.py |
Main orchestration entrypoint |
route |
src/llm/router.py |
Tool selection based on prompt intent |
MemoryStore.save |
src/memory/store.py |
Prompt and response persistence |
trace |
src/observability/tracer.py |
Structured event emission |
Routing and execution decisions
| Situation | Recommended action | Reason |
|---|---|---|
| Tool intent is obvious | Route deterministically | Fast, explainable behavior |
| Tool intent is ambiguous | Ask clarification or fallback | Prevents wrong side effects |
| Tool fails repeatedly | Trigger fallback and alert | Maintains UX and reliability |
| Memory grows noisy | Apply retention and relevance rules | Keeps context quality high |
Monitoring cheat cards
| Metric | Healthy signal | Warning signal |
|---|---|---|
| Route success rate | Stable and high | Drops after release |
| Tool error rate | Low and bounded | Spikes during traffic peaks |
| P95 latency | Within SLO | Sustained increase over baseline |
| Fallback rate | Low and expected | Rising due to routing drift |
What should I check first in an incident?
Start with route decision logs and tool error metrics. Most user-visible failures originate from those two layers.
What single function should every engineer understand first?
handle_prompt.
It is the control point where behavior, quality, and safety meet.
--8<-- "_abbreviations.md"