Pricing, evaluation, security, and what happens to your data.
25 questions
Plans & billing
One traced span, typically one LLM call, one retriever call, or one decorated function invocation. A single end-to-end agent run usually emits 5-20 traces depending on how many tools and LLM calls it makes. Trace ingestion is unlimited and free on every plan. The Free plan keeps a rolling 14-day history, while Team and above retain your traces indefinitely.
Plans & billing
One LLM-as-judge scoring call: e.g. a hallucination check on an answer or a relevance score over a retrieved chunk set. Metrics include hallucination, faithfulness, relevance, toxicity, coherence, and completeness. Free includes 100 evals/month, Starter 2,000, Team 10,000, Growth 50,000, and Enterprise is unlimited. Beyond your allowance you pay per evaluation, priced by depth: $0.007 for a single-judge metric up to $0.545 for a deep agentic run with a multi-model jury, because those are genuinely different amounts of work.
Plans & billing
Your plan includes a monthly allowance of evaluations and security scans. Beyond it you pay per evaluation, priced by depth: $0.007 for a single-judge metric, $0.015 for a fast agentic run, $0.065 for a standard one, and $0.545 for a deep run with a multi-model jury. Security scans are $0.0005 each. Every price includes the judge tokens. You can set a hard spend cap so a runaway batch cannot surprise you.
Plans & billing
Because it is genuinely more work. A fast run does deterministic checks plus one judge call. A deep run reads the entire trajectory, then convenes a jury of several models and keeps each juror's score and reasoning as an audit trail. That can be a hundred times the judge tokens of a single relevance check. Most tools bill both as one evaluation, which means light users subsidise heavy ones. We would rather show you the difference.
Plans & billing
Yes, on every plan including Free. Save a provider key in Provider Keys and your evaluations run on your account, so judge tokens bill to whatever rate you already negotiated instead of to us. Keys are encrypted before storage with a data key we cannot read without AWS KMS. After you save one it is never shown again, never sent to your browser, and never written to a log. If a key stops working the evaluation stops rather than quietly falling back to our account.
Plans & billing
No. Plans include seats and you are never billed by headcount. Adding a teammate who needs to look at one trace should not cost the same as adding a heavy user, and per-seat pricing tends to stop people sharing the thing you want them to share.
Plans & billing
Evaluations and security scans stop for the rest of the month; tracing keeps running, because tracing is unlimited on every plan and we do not want your observability to go dark over a billing threshold. You can upgrade at any time, or start a five-day trial of a paid plan without a card.
Evaluation
Yes, on every plan including Free. Pick the judge model per run, per agentic evaluation, or per dataset batch. At deep depth you also choose the jury: which models sit on the panel, each one scoring independently with its reasoning kept on the result. Different judges disagree in interesting ways, and seeing that disagreement is often more useful than a single confident number.
