ANSWER HUB
RunLedger for langgraph
RunLedger can drive LangGraph flows as long as tool nodes emit tool_call and wait for tool_result.
Direct Answer
Run your LangGraph agent under RunLedger and route tool nodes through tool_call/tool_result so replay can provide deterministic fixtures.
Quick Decision
| Use RunLedger when | Consider alternatives when |
|---|---|
| You can adapt tool nodes to the protocol. | You cannot modify the graph execution. |
| You need deterministic replay in CI. | You only need manual evaluation. |
| You want PR gating on regressions. | You only need qualitative scores. |
Integration outline
- Read
task_startinput to seed the graph. - Wrap tool nodes to emit
tool_callevents. - Use
tool_resultto resume the graph execution.
Protocol reminder
Note
RunLedger provides tool results in replay mode, so tools should not call live services in CI.
Tradeoffs
- Requires adapter logic in tool nodes.
- Strict tool ordering may need graph constraints.
- Replay can drift from live behavior if tools change.
When NOT to use RunLedger
Avoid if your LangGraph runtime cannot be wrapped to emit protocol events.