AI Privilege & Quality Settings
Enable AI for your organization and tune quality levels per feature.
The AI Credits page (Organization → AI Credits, /admin/ai-credits) is the control panel for AI features across your organization. From one page you can:
- Toggle AI features on or off for the whole org.
- See your available credits and this month's usage at a glance.
- Pick an AI quality level per feature (AI Agent vs. Test Case Generation).
- Browse a credit usage history of every AI call.
Who can access this page
Every org member can open the page and read the balance. Only the organization owner can change the Enable AI toggle or the per-feature quality levels. The controls are shown to everyone, but anyone other than the owner who tries to change one gets an error and the control snaps back.
Page layout
At the top of the page, three cards summarize status:
| Card | What it shows |
|---|---|
| Enable AI Features | On/off toggle. Disabling it stops AI calls across the org — AI Agent and AI test case generation. |
| Available Credits | Credits left to spend — the remainder of this month's allowance on a paid plan, or of the one-time grant on a free organization. |
| This Month Usage | Credits consumed this month across all AI features. |
Below the cards, two collapsible sections:
- AI Quality Settings — per-feature quality level (covered below).
- Credit Usage History — paginated table of every AI call, who ran it, which feature, and credits consumed.
Enable AI Features
Flip this toggle off if you want to temporarily disable all AI calls — for example, during a billing review or while training a new team. Effects:
- AI Agent won't open.
- AI test case generation button is disabled.
- Insight reports and the score badges are unaffected — they use no AI and no credits.
Existing data (previously generated test cases, past AI agent conversations) is unaffected.
AI Quality Settings
Each AI feature has an independent quality level. Pick based on the trade-off between speed, cost, and output quality.
The rate is set per feature, so the AI Agent and Test Case Generation can sit on different levels. The figures below are how many tokens one credit buys at each level — the higher the number, the cheaper the level.
| Level | Speed | AI Agent | Test Case Generation | Use when |
|---|---|---|---|---|
Basic (cheap) | Fastest | ~34,000 tokens per credit | ~22,500 tokens per credit | Quick iteration, bulk generation of straightforward cases |
Balanced (medium) | Default | ~8,500 tokens per credit | ~3,000 tokens per credit | Everyday use — sensible default for most projects |
Premium (best) | Slowest | ~2,850 tokens per credit | ~2,850 tokens per credit | Complex specs, critical flows, when you want the best reasoning |
Both features start on Balanced.
The quality level is the biggest single lever on how fast you spend credits: the same request costs roughly 12 times more on Premium than on Basic for the AI Agent, and roughly 8 times more for test case generation. Check this setting before concluding you need more seats. See What an AI Action Costs for how a charge is calculated.
AI Agent level
Applies to the conversational AI Agent — sidebar chat (Ctrl+K). See AI Agent.
Test Case Generation level
Applies to the Create with AI action in the Test Cases toolbar. See Test Generation.
Changing a level
Click the level button (Basic / Balanced / Premium) for the feature you want to change. The new level saves immediately and a toast confirms; failures revert to the prior level.
Credit Usage History
Each row in the history table shows:
- Date — when the call happened, in your local timezone.
- User Name — who triggered the AI call.
- Use case — which feature used credits (e.g., "AI Agent", "Test Case Generation").
- Credits — how many credits the call consumed.
Use this to audit unusual spikes — for example, if your credit balance drops faster than expected, check the history for unfamiliar usage.
Monthly History
The same page carries a Monthly History section with credits granted, consumed and expired per billing month. See AI Credits (org billing).