LangSmith is an excellent trace viewer and dataset manager for LangChain apps. Its playground can re-run a captured LLM call, and LangGraph can fork a run from a checkpoint — but the fork re-executes tools live, and nothing produces a signed audit record.
| Capability | LangSmith | Runback | Note |
|---|---|---|---|
| Read and search traces | ✓ | ✓ | |
| Dataset and eval management | ✓ | ✓ | |
| Re-execute any step from captured inputs | partial | ✓ | LangSmith's playground re-runs any captured LLM call with edits, and LangGraph time travel forks a LangGraph run from a checkpoint — but the nodes after the fork re-execute tools and API calls live. Runback replays the step with the captured tool and environment results held fixed, so the re-run is deterministic |
| CI release gate — policy blocks the deploy | partial | ✓ | pytest, Vitest and Jest integrations run LangSmith evals in CI; there is no packaged gate Action |
| Signed, tamper-evident audit export | ✗ | ✓ | SHA-256 chain; verify without Runback |
| Self-host, data never leaves your perimeter | partial | ✓ | A full deployment exists, but it's Enterprise-tier only, license-gated, and some features (Hub, cloud-model evaluators) stay cloud-only |
| Regulatory mappings (EU AI Act · CPS 230 · NIST) | partial | ✓ | Publishes an EU AI Act capability crosswalk; no CPS 230 or NIST AI RMF mapping |
| Redact PII/secrets automatically, on by default | partial | ✓ | LangSmith's Python/TS SDKs ship a create_anonymizer() toolkit (regex, custom functions, or Presidio/Comprehend) — capable, but nothing is redacted unless you write the patterns yourself, and it isn't available in LangSmith's Go SDK. Runback's TS, Python, and Go SDKs all redact in-process with working detectors on by default |
| One flat price — no separate compute/storage units | ✗ | ✓ | LangSmith bills seats, base traces and usage units (LSU, $1.00/unit), plus a paid upgrade to extended (180-day) trace retention. Runback is one tier: seats + runs + retention, bundled |
| Purpose-built Go SDK, not just a generated REST client | partial | ✓ | LangSmith's Go package is Stainless-generated REST API bindings plus a bare OpenTelemetry tracer — no traceable()-style ergonomics and no anonymizer. Runback's Go SDK has the same manual-recording shape (and redaction) as its TS/Python SDKs |
Its playground can re-run a captured LLM call, and LangGraph can fork a run from a checkpoint — but the fork re-executes tools live, and nothing produces a signed audit record. Most teams use both — LangSmith for what it's already good at, Runback for deterministic incident replay and the signed record.
Source: LangSmith's public documentation, https://docs.smith.langchain.com/, read 8 October 2026. Products change: report an error to legal@runback.dev and we will correct it. Names and trade marks belong to their owners; Runback is not affiliated with, endorsed or sponsored by any of them.
No signup. Walk the scripted failing demo run step by step and see what the model was given.