Simple enough to approve, honest about the limits.

Every plan brings your own model keys. What you pay for is the evaluation workspace, collaboration and evidence, not the tokens your evaluator or judge consumes.

Free
$0

Forever. No card required.

  • Seats: 2
  • Projects: 2
  • Eval cases: 1 000 / mo
  • Runtime traces: 5 000 / mo
  • Trace history: 14 days
  • Voice call minutes: 30 / mo
  • Single-turn and multi-turn text eval
  • 5,000 welcome credits to run your first evaluation
  • Platform credits: bought separately, same packs on every plan
Start free
Recommended
Team
$99/month

Billed monthly. Cancel anytime.

  • Seats: 5
  • Projects: 10
  • Eval cases: 10 000 / mo
  • Runtime traces: 25 000 / mo
  • Trace history: 180 days
  • Voice call minutes: Unlimited
  • Red teaming on the platform model, paid in credits
  • Platform credits: bought separately, same packs on every plan
Start free, upgrade later
Custom
Custom

For your perimeter and process.

  • Seats: Unlimited
  • Projects: Unlimited
  • Eval cases: Unmetered
  • Runtime traces: Unmetered
  • Trace history: Your instance
  • Everything in Team
  • Workspace roles and per-project access control
  • Audit log of every action, exportable

Answers before you invite finance.

Something not covered here? Book a demo. We'll walk through billing, quotas and the enterprise plan.

  • Do you charge for the tokens my evaluations consume?
    Only when you run on the platform model. Bring your own provider key and every target, judge and generation call is billed by that provider straight to you; the subscription pays for the workspace, not the inference. Prefer not to manage a key? Run on the platform model and pay in platform credits instead. It is a per-workspace choice you change in Settings, and every run records which model scored it.
  • What are platform credits, and what do they pay for?
    Credits are prepaid units that cover the work you could otherwise run on your own key when the platform model does it for you: judge scoring, dataset and case generation, the multi-turn user simulator, trace analysis and the voice agent. AI writing hints and the red-team attacker do not spend credits; they come with the subscription. Every new workspace starts with 5,000 welcome credits, and you top up with packs at any time, the same packs on Free and on Team. No plan includes a monthly credit allowance, so you only buy more if you actually run on the platform model.
  • Is red teaming included, or a paid add-on?
    Included on every plan, with no separate purchase. Free covers 200 attack cases a month and Team covers 2,000, where one case is a generated single-turn attack or one emulated multi-turn session. The adversarial attacker runs on the platform's own model, so there is no key to bring for it, and an attack success rate can gate a release in CI like any other metric.
  • What counts as one evaluated case?
    One dataset row scored by one run, no matter how many metrics you attach to it. A 500-row dataset run twice is 1,000 cases. Judge retries do not add to the count, and a run that fails before scoring, from a bad connector or a cancel, is not charged at all.
  • What happens when I cross a quota?
    Nothing stops mid-run. An eval run that crosses the line finishes rather than dying half-way through a large dataset, the overage lands on your next invoice ($0.02 per eval case, $0.09 per voice minute, $1.50 per 100,000 traces), and we email you at 80% and 100% so it is never a surprise. If you would rather block than spend, hard caps are available.
  • How long do you keep my data?
    Your own work, the datasets, test plans, connectors, reports and the score history of every run, is kept for as long as your workspace exists, on every plan including Free. What expires is the raw record underneath: runtime traces after 14 days on Free and 180 on Team, and the per-row detail of a run after 30 days. Charts, run scores and comparisons keep working across your whole history; what you lose is opening one old row to read the model's exact answer. Custom runs on its own instance, where retention is set per deployment.
  • What happens to my history if I downgrade?
    Nothing for 30 days. Moving from Team to Free gives you a month to export anything the shorter window would drop, and we email you before the first deletion. Datasets, reports and run scores are never deleted on any plan.
  • Do you offer a free plan and self-hosted?
    Both. Free is permanent, with no card and no trial clock: 2 seats, 2 projects, 1,000 eval cases and 5,000 runtime traces a month, red teaming and 30 voice call minutes, plus the 5,000 welcome credits. Self-hosting is part of Custom, where the same service set deploys inside your perimeter with your own Postgres, ClickHouse and object storage, licensed per environment.

    • Start on the Free plan. No card, no clock.

    Start free