Talk to sales

Great for building agents. Not built for auditors.

Langfuse and LangSmith help engineers trace, evaluate and improve agents. Runfile keeps a sealed record of what those agents did, for the people who have to answer for it.

The short answerKeep them for building and evaluating. Add Runfile for the record.

See them side by sideBook a design-partner call
01 — Different jobsUse both

Two tools, two different jobs.

What each is built for, and when to reach for which.

Langfuse · LangSmithLLM engineering platforms

Built to iterate.

“Is this prompt or model change better?”
Bought by
AI engineering
Records
Traces, evals, datasets and prompt versions
Kept for
Depends on plan or your own database
RunfileThe agent system of record

Built to prove.

“What exactly did this agent do, and can you prove it?”
Bought by
CCO · Internal Audit
Records
The agent execution graph, sealed
Kept for
Up to 7 years, on locked storage

Choose Langfuse or LangSmith for

  • Tracing while you build and debug
  • Evals, datasets and prompt management
  • An open-source, self-hosted stack (Langfuse)

Choose Runfile for

  • A record intact since it was written, not since your last redeploy
  • Evidence mapped to the rules you answer to
  • A package your auditor can verify offline
02 — Head to headHonest verdicts

Side by side, honestly.

Eight things teams ask about. Where Langfuse or LangSmith is the better choice, we say so.

What you needLangfuse & LangSmithRunfile
Trace agents while you buildYesTheir core job.NoNot built for iteration. Keep them for this.
Evals, datasets and prompt versionsYesBuilt in.NoOut of scope for Runfile.
Self-hostingYesLangfuse is open source; LangSmith offers self-hosting for enterprises.PartlySingle-tenant today; in your own cloud from v1.5.
Proof the record hasn’t changedNoTraces sit in an ordinary database, with no proof they’re unchanged.YesHash-chained, signed daily and anchored in a public log.
Kept for years, on locked storagePartlyRetention depends on your plan or your own database.YesUp to 7 years on S3 Object Lock.
Each event linked to the rule it satisfiesNoNo link between events and controls.YesMapped to the frameworks you answer to.
A package to hand the auditorNoBuilt for engineers, not auditors.YesSigned PDF, raw event log and verifier, together.
Checked by an outsider, offlineNoNo independent verification.YesAn offline verifier and a public anchor.

Based on each vendor’s public product information. Products change, so check with them for current details.

03 — Better togetherNothing to rip out

Run them side by side. Change nothing.

Runfile runs on the same agents your engineers already trace. Their workflow stays the same.

StartYour AI agentsLangGraph, OpenAI Agents SDK or any OpenTelemetry source
Traces
UnchangedLangfuse or LangSmithEngineers keep tracing, evaluating and iterating.
Same runs
AddedRunfileKeeps the sealed record your auditor checks.

Questions about this comparison

Why not run Langfuse self-hosted with longer retention?

Self-hosting solves where the data lives. It doesn’t solve whether it can be changed, how it maps to controls, or whether anything is missing. A regulator wants to know the log is intact since it was written, not since you last redeployed.

Can’t we add a hash chain ourselves?

You can, and several teams have. Then you need signing keys you don’t control, a public anchor, locked storage, control mapping and a verifier your auditor trusts. That’s the product, and those teams usually come to us next.

We use LangGraph. Does that change anything?

LangGraph is supported natively, so Runfile runs next to LangSmith on the same agents.

Other comparisons

CompareRunfile vs DatadogCompareRunfile vs Vanta & DrataOverviewAll four kinds of tool
§ Next step

See it running
next to Langfuse or LangSmith.

A 30-minute call with your AI engineering and Internal Audit leads. We’ll show Runfile running next to your tracing tool on the same agents.

Book a design-partner callSee a sample evidence package