How-to guidesΒΆ
Task-oriented guides. Each one starts from a working graph and shows the smallest change that solves the task, with real output.
| Guide | Use it when you need to |
|---|---|
| Explain a decision after the run | Explain decisions once invoke has returned, or later from persisted records |
| Handle human-in-the-loop interrupts | Record pauses, human answers, approvals, and the checkpoint to replay |
| Correlate multi-agent runs and retries | Follow one request across nodes, agents, retries, and services |
| Capture MCP tool calls | Record tools served by a Model Context Protocol server |
| Generate LLM-phrased explanations | Phrase explanations in natural language with a chat model |
| Test disclosure policies | Prove that each audience sees exactly what it should |
| Configure failure modes | Decide whether instrumentation failures may affect your graph |