Credits & Billing

Agent Bayes meters AI usage in credits. This page explains where credits come from, what consumes them, and how to track usage.

Where credits come from

There are two kinds of credits in your balance:

• Recurring credits: granted every month by your plan. On a paid plan they expire 2 months after they are granted, so one month of unused credits rolls over. The Free plan grants 50 credits every month from the day you verify your email, and they expire when the next month's credits arrive.
• Add-on credits: purchased separately as a top-up. They expire 6 months after they are granted.

Credit consumption begins with the oldest recurring credits in your balance. Once all recurring credits are exhausted, add-on credits are consumed.

Both kinds live in the same balance as separate pools, and the Credits page in Settings shows them separately so you can see what will expire and when.

What consumes credits

Credits are spent on AI operations. The main types you'll see in your history:

• Mindmap — agent workflows.
• Indexing — processing an uploaded PDF.
• AI Verification — independent verification passes on claims and citations.
• Translation — translating the original text of search results.

All of the above actions are manually triggered by you and are metered in 0.1-credit increments.

What things cost

Roughly, on recent measurements:

Indexing cost is fairly steady per page. Agent runs vary a lot, because the agent decides how many searches and how much writing a question deserves. The model you pick sets the upfront reservation, not the final price, so a cheaper model on a hard question can still cost more than an expensive model on an easy one.

Reservations and refunds

Before an operation starts, the system reserves an estimated cost from your balance. A research agent run reserves 50 credits times the model's multiplier: 5 on GPT-6 Luna, 20 on Gemini 3.8 Flash, 45 on GPT-6.1 Sol, 60 on Gemini 4 Argon, 140 on Claude Opus 5.5, and 225 on GPT-6 Astra. Indexing reserves 0.5 credits per page, which is deliberately above what it usually costs, so a long or unusually dense document cannot run out part way through. When the operation completes (or you cancel it), the reservation is released and you are charged for what was actually used, which is normally less than was held. This is why your balance can briefly dip and then partially recover.

If a reservation can't be made because your balance is too low, the operation won't start, and you'll see an error explaining why.

Buying top-ups

When you need more credits before your next billing cycle, you can buy an add-on credit package in Settings → Credits. Three packages are available:

Pick a package and pay with PayPal. The credits land in your add-on pool right after the payment is confirmed, and they expire 6 months after purchase. Add-on credits are consumed only after your recurring credits, so a top-up never shortens the life of your monthly allowance.

If you find yourself topping up regularly, compare with the next plan tier first. Plans carry a lower per-credit rate, so upgrading is usually the cheaper way to get more credits every month.

Viewing your balance and history

Settings → Credits shows:

• Your current balance at the top.
• A full credit history with one row per grant, reservation, or usage entry, including operation type, timestamp, and the credit delta. Top-up purchases appear as their own entry type, with the expiry date next to each grant.

Running out

If your balance hits zero:

• New AI operations (instructions, indexing, verification) will be blocked.
• Existing mindmaps, libraries, and documents remain accessible — nothing is deleted.
• Buy a top-up package in Settings → Credits, or wait for next month's recurring grant.

What's next

• The Interface
• Confidence, Gaps & Iteration Limits — how the 3-loop cap protects your budget.