Rewind
Roll a running or failed workflow back to a previous checkpoint. The run state resets to the checkpoint moment: completed steps are preserved, subsequent steps are cleared, and execution resumes from that point. Useful for recovering from failures without restarting from scratch.Branch
Create a new run from a checkpoint with modified inputs. The original run is unaffected. The branched run starts from the checkpoint state with new inputs applied. Useful for “what if” exploration and A/B testing of workflow variations.State Preservation
Each checkpoint stores:| Field | Description |
|---|---|
completedStepIds | Steps that finished successfully up to this point |
skippedStepIds | Steps conditionally skipped up to this point |
variables | Full accumulated variable scope (args + step outputs) |
accumulatedOutputs | Per-step output objects keyed by step ID |
variableOverwriteOrder | Step ID sequence for deterministic replay |
sequence | Monotonic counter within the run |
Checkpoint state uses the same structure as
failureSnapshot. Rewind and restart share the same state restoration path in the Convex interpreter.How It Works
Step 1->Checkpoint A->Step 2->Checkpoint B->Step 3 FAILED
Rewind to A -> re-execute steps 2, 3 from checkpoint A stateBranch from B -> new run with modified inputs, original run unchanged
In Burgundy: Manage checkpoints from the run detail page. ->