Skip to content
TestAutomate Join the waitlistWaitlist

Get notified at launch

TestAutomate isn't released yet. Leave your name and email and we'll notify you when it launches.

Home / Product / Economics ECONOMICS

BYOK AI testing: model spend at cost

Pay what the model costs. Nothing on top. Bring your own AI key and model spend passes through at cost — with a live meter on every run and a hard budget ceiling you set. No credits to decode, no markup to guess at.

BYOK: your key, your provider price

Most AI testing tools resell model usage as "credits" at a conversion rate you can't audit. TestAutomate doesn't. Plug in your own key and every token is billed by your provider, to you, at your price. We show you the dollars as they accrue.

  • Keys supported today. Anthropic, Amazon Bedrock, Google Gemini, and Groq. OpenAI and xAI keys are on the roadmap. coming
  • At-cost pass-through. Model spend is your provider's price — TestAutomate adds nothing on top of BYOK usage.
  • Live meters. Every run and every sprint sweep shows its spend in dollars, updating as it goes, with the model named.
  • Hard ceilings. Set a budget on a run or a sweep. This is a product promise: when spend reaches the ceiling, the work stops cleanly and reports what it graded so far. Nothing keeps spending in the background.

Cheap by engineering, not by discount

At-cost billing only matters if the cost itself is low. TestAutomate is built to spend as few tokens as possible per verdict.

HOW 01

Accessibility tree first

The agent reads your page's accessibility tree — structured text — instead of pushing a screenshot through a vision model at every step. Text tokens are a fraction of the price of image tokens. Screenshots are still captured where they belong: as evidence in your verdicts.

HOW 02

Model tiering

An inexpensive model drives the browser for routine steps. Stronger, pricier models are reserved for the moments that earn them — grading verdicts and re-driving failures on escalation. You pay for judgment, not for clicking.

HOW 03

Caching and learned skills

Login recipes, nav notes, and playbooks from your docs mean later runs skip the exploration earlier runs paid for. Prompt caching cuts the repeated context cost. The tenth run of a test is meaningfully cheaper than the first.

Credit-metered pricing vs. at-cost pricing

Credit-metered AI SaaSTestAutomate
What a run costsSome number of "credits" — the dollar conversion is opaque and set by the vendorProvider price, passed through at cost, shown live in dollars
When models get cheaperVendor margin widens; credit prices rarely followYour cost drops the day your provider's price does
VisibilityA credit balance draining at an unstated ratePer-run and per-sweep meters, model named, dollars stated
OverrunsA surprise top-up, or a hard credit wall mid-runHard budget ceilings you set — work stops cleanly at the line
Your testsLocked in a proprietary formatPortable, human-readable YAML you can take anywhere

Three ways to pay — pick per team

LANE 01

Bring your own key

Anthropic, Bedrock, Gemini, or Groq today. At-cost pass-through, live meters, hard ceilings. The default for teams that already have a provider relationship.

LANE 02

Managed credits

No key? Buy prepaid credits and start immediately. You pay the provider's price plus a stated fee — stated, as in printed next to the price, not buried in a conversion rate. Switch to BYOK any time.

LANE 03

Your Claude Code subscription

Developers can run TestAutomate locally on the Claude Code subscription they already pay for — zero marginal model cost per run. The whole stack self-hosts on Node and Postgres. See the developer workflow.

The meter numbers on this page are illustrative UI, not benchmarks. Your spend depends on your app, your tests, and your provider's prices — which is exactly why the meter exists.

Set a ceiling. Watch the meter. Keep the difference.

Start with your own key or with credits — either way, you'll always know what a verdict cost.

Further reading