Billing and plans
Free, Team, and Custom plans; how upgrades and the Stripe portal work.
Last updated 2026-09-17
Billing is workspace-scoped, and the workspace owner is the only person who can change it: starting or resuming a subscription, and opening the Stripe portal where the saved cards and invoices live, are owner-only. An admin can buy credit packs. Everyone else sees the plan, the usage meters and the credit balance, and changes nothing. All of it lives on /settings?section=billing.
Plans
| Plan | Projects | Seats | Eval cases / mo | Runtime traces / mo | Support |
|---|---|---|---|---|---|
| Free | 2 | 2 | 1 000 | 5 000 | Community |
| Team ($99/mo, $948/yr) | 10 | 5 | 10 000 | 25 000 | Email, next business day |
| Custom | Unlimited | Unlimited | Unmetered | Unmetered | Named engineer, shared Slack |
Free is permanent, not a trial: the quotas are bounded but nothing expires.
Team adds two features Free does not have: voice evaluation (outbound calls to your bot, with transcript and scoring) and red teaming (10 vulnerability classes, 9 attack techniques). Both are visible on Free with a Team chip, and picking one points at the upgrade instead of failing at save time.
Custom adds self-hosted deployment, SSO / SAML, RBAC beyond the default roles, audit-log export, per-deployment data retention, and a signed DPA. Contact sales for a quote.
What you're billed for
- Platform features: the plan tier.
- Model provider calls: not billed by EvaliQA. They land on your provider invoice via BYOK.
Platform credits
Every workspace has a wallet of platform credits. Credits pay for the AI work you could otherwise run on a key of your own, while the Platform AI agent runs on the platform model: LLM-as-a-Judge scoring, dataset generation, the adaptive simulator, trace analysis and the voice agent. You never see or pick the model; you see credits. The small in-app AI hints (Suggest goals, Suggest metrics) and the red-team attacker are the platform's own spend, counted against your plan's allowance, and cost no credits on either source.
- Price. Every call on the platform model is charged in credits, depending on the model tier and how much text it reads and writes. A typical judge call costs tens of credits; a 50-row run with four metrics, a few thousand. The balance and the history on Settings → Billing are in credits only.
- Where credits come from. An account gets 5,000 welcome credits once, in the first workspace it owns. Once per account rather than per workspace: the credits are a welcome to the platform, not a reward for making workspaces. Somebody who joined by invite and later leaves to start their own workspace gets them then. Packs (25,000 / 100,000 / 500,000) can be bought at any time by an owner or admin from Settings → Billing, on Free and on Team alike. Credits never expire. No plan includes a monthly credit allowance: the subscription buys wider access to the product (seats, projects, quotas, voice), credits are always bought separately.
- When they run out. A platform-model call is refused with a No credits left message. Nothing else stops: buy a pack, or switch the Platform AI agent to your own key. Custom and Internal plans are not metered in credits.
- Credits or your own key. The choice is made once, at sign-up (Use my bonus credits or Use my own LLM key), and lives in Settings → Platform AI agent, nowhere else. On credits, no LLM key is asked for anywhere: the Get started checklist has no credential step, Quick Eval has no judge step and the test plan wizard has no LLM Judge step. Switching to your own credential shows a warning first: judge scoring and every AI operation above then run on your key at your expense, and results may vary with the models you pick. From then on the credential and judge pickers come back in every flow.
The header shows the balance; Settings → Billing shows the balance and the last transactions.
Upgrade
Settings → Billing → Upgrade, monthly or annual. Checkout opens on Stripe and returns you to the same page; access changes only once Stripe's signed webhook confirms the payment, which is usually seconds and is why the page says the payment is processing in the meantime.
Stripe portal
The Manage billing button opens the Stripe customer portal in a new tab, for the owner only: it shows the saved cards and every invoice ever paid, which is not something an admin or a platform administrator gets to see on the owner's behalf.
From the portal you can update the payment method, download invoices and cancel the subscription.
Cancelling, and changing your mind
Cancelling in the portal does not end anything on the spot. The subscription keeps working to the end of the period you have paid for, and Settings → Billing says so: Cancels Oct 17, 2026 instead of a renewal date, with a Resume subscription button next to it. Resuming calls the cancellation off and the plan carries on as before.
A workspace has one owner and one subscription: deleting the workspace cancels the subscription as part of the deletion.
If a payment fails
A refused renewal does not lock the workspace immediately.
- The plan and its limits stay. Stripe retries the card on its own schedule, and the page names the day it will try again, taken from Stripe rather than guessed.
- New runs, projects and invites keep working for 7 days after the first failure, to leave room to fix the card.
- The banner asks for the payment method, not for an upgrade: the plan is not the problem.
- If Stripe exhausts its retries, what happens next is a Stripe account setting: it can cancel the subscription, park it as unpaid, or leave it past due. Cancellation drops the workspace to Free, keeping the data under Free's retention windows.
Data retention
How long recorded data stays readable depends on the plan. Two clocks, because the two kinds of data lose their value at very different rates:
| Runtime traces | Row-level eval results | |
|---|---|---|
| Free | 14 days | 30 days |
| Team | 180 days | 365 days |
| Custom | Unlimited | Unlimited |
What is never deleted, on any plan: your projects, datasets, test plans, connectors, custom metrics, reports, and every eval run with its summary scores. Metric trends and run comparisons keep working over your whole history no matter how short the window is. What expires is the detail inside a run — the inputs, the model's answers, the judge's verdicts, row by row — and the runtime traces.
Getting your data out
Both kinds of data export before they expire:
- Traces — the Export CSV button on the Tracing page. It exports everything matching the filters you have set, including the full spans and metadata, not just the columns on screen.
- Eval run results — the CSV button on a run's page: one row per case, with every metric's score, verdict and reason.
What you are told, and when
- The first time anything in your workspace approaches its window, the workspace owner gets one email. Only the first time — after that something crosses every day, and a daily email about it would be noise.
- If a plan change shortens the window, the owner is told at the time of the change. Nothing is deleted for 30 days after a downgrade, so there is time to export or to change your mind.
Enterprise self-hosted
The Enterprise plan includes an on-premises deployment kit: a Kubernetes Helm chart, an Alembic migration bundle, and a version-lock policy that ships updates in 30-minute cutovers rather than blue-green rollouts. Contact sales.
Closing an account
Settings → Profile → Delete my account schedules the deletion 30 days out, and the same panel calls it off until then. Every workspace where you are the only member goes with the account, and its subscription is cancelled with it. If you are the only owner of a workspace other people use, hand it over or delete it first: the panel names the workspaces that are in the way.