← Vetro Attest · Free Art. 12/14 checker · Free EU AI Act tools

Vetro Attest vs. Braintrust

*Engineering comparison, not legal advice. Braintrust features change quickly; check their documentation for the current state.*

Braintrust is where many teams decide whether a new prompt or model is better than the old one. Vetro Attest is where a single run that mattered becomes evidence someone else can check. One is about quality over many runs, the other about proof for one run.

What Braintrust does well

Braintrust is a commercial platform for evaluating and monitoring AI products. Teams use it to:

• Run evals: score a prompt, model or agent version against a dataset with code-based, human or LLM-as-a-judge scorers, and compare experiments side by side;
• Log production traffic as traces with spans for LLM calls and tools, and turn interesting logs into dataset rows;
• Iterate on prompts in a playground linked to the same scorers;
• Monitor quality, cost and latency over time.

For the loop "change something, measure whether it got better", it is a strong choice.

What an eval and logging platform does not give you

Evidence for a regulator, an auditor or a customer must convince someone who does not trust you.

• The operator controls the data. Logs live in a project you administer. You, or anyone with enough access, can edit or delete them, so an export is a claim, not proof.
• Aggregate scores are not per-run evidence. "Accuracy went from 91% to 94%" is useful for engineering; a complaint about one decision needs the record of that one decision, intact.
• No independent timestamp or signature. An export shows what the project held when you exported it, not that it looked the same when the decision was made.
• No mapping onto the AI Act. Logging duties (Art. 12 for providers, Art. 19 and 26 for retention) ask what the logs make traceable, not only what they contain.

None of this is a flaw in Braintrust; it is outside what an eval platform is for.

What Vetro Attest adds

Vetro Attest accepts an OTLP/JSON trace with the OpenTelemetry GenAI semantic conventions, or a natively signed Vetro ledger, rebuilds it into a hash-chained ledger and checks it for free. For 49 USD, one-off, you get:

• a full EU AI Act report (Markdown and HTML) mapping the trace onto logging obligations;
• an ed25519 countersignature over the digest, event count and timestamp, verifiable by anyone against our published public key, no account needed.

The countersignature attests that the trace verified at a time. It does not attest that the events are true; the report is a structured aid, not legal advice or certification.

When to use both

• Keep Braintrust to evaluate versions and watch quality in production.
• When a single run matters, such as a refusal, an approval or a flagged output, export its trace as OTLP/JSON and get it countersigned. Store the attestation next to the trace.
• On an audit or dispute, hand over the trace plus the attestation; your eval history shows diligence, the attestation shows the record was not changed.

Braintrust tells you whether your agent is getting better. The countersignature lets someone else believe what it did once.

Try it

• Run the free EU AI Act Art. 12/14 checker.
• Paste an OTLP trace on the Vetro Attest home page and verify it free. The full report plus countersignature is 49 USD, no subscription.

See a sample ledger verified in one click, check your own logs free or get the countersigned EU AI Act report.