Platform/Monitoring

Replay, compare, reproduce.

Every run records its inputs, its tool calls, its LLM completions, its branch decisions, and its outputs. Replay any run from any node. Compare two runs side-by-side. Debug a graph the way you debug code.

What you get

Six observability primitives.

01 · Run timeline

Every node, every ms.

Latency, status, cost, retries — per node. The graph turns into a Gantt chart on hover.

02 · Replay

From any node forward.

Pick a node. Edit its inputs. Re-run downstream only. Original run preserved alongside.

03 · Compare

Two runs, side-by-side.

Diff inputs, outputs, branch decisions, tool calls. Find the one node that changed behavior.

04 · Trace

OTel-compatible spans.

Every node emits a span. Pipe to Datadog, Honeycomb, Grafana — your existing observability stack.

05 · Cost

Per-run, per-tenant, per-node.

Token cost, tool cost, compute cost. Roll up to workflow, tenant, or workspace.

06 · Audit

Who, what, when.

Every form decision, prompt edit, version pin. Immutable record. SOC 2 / HIPAA ready.

A run timeline

What a single run looks like.

RUN · onboarding · 04f7-acme · v3.2 COMPLETED · 4.2s · $0.014
+0mswebhook · signup_received3 msok
+12mstool · kyc.run812 msok
+830msllm · risk_score (haiku-4.5)1.4 sok
+2.2sbranch · score > 0.7 → manager1 mstrue
+2.2sform · manager_review1.9 sawaited
+4.1stool · provisioning.create112 msok
+4.2snotify · slack38 msok
Debug like code

Bring a flaky run. We'll replay it.

30 minutes. Hand us a workflow that intermittently fails. We replay it, find the diverging node, and fix it together — live, on your data.