Credits

Credits are the shared unit Foundation Machines uses to price AI-backed work. For plan pricing, see billing. To start from installation, see the quickstart.

Why credits, not tokens

Model providers price in tokens, but tokens are a provider-specific implementation detail. Credits give customers one stable unit across PR reviews, chat replies, autofix runs and future Foundation Machines products.

That abstraction lets us change model providers, add new models or tune model blends without making you learn every provider's token SKU. You see credits and dollars; we handle the model math underneath.

The model rate card

Credits map to model input and output tokens through an internal rate card. The current default blend is Claude Sonnet 4.5 plus Claude Haiku 4.5: Sonnet handles higher-reasoning review work, while Haiku handles cheaper supporting tasks where it is enough.

When we add new models or change the blend, the rate card adjusts so the customer-facing price stays stable. One credit remains $0.02 on the Team plan even if the underlying provider prices change.

Reading a receipt

Per-PR receipt comments are off by default. When enabled, a receipt shows the model, token counts, credits and dollar equivalent for the review:

Sebastion review · model: claude-sonnet-4.5 · input 78k / output 9k tokens · 64 credits ($1.28)

Use it to spot unusually expensive PRs, calibrate your spend cap and understand how large refactors differ from small changes.

Opting in to receipts

Add this to sebastion.yaml in the repository:

billing:
  receipts: true

Receipts are opt-in so normal PR threads stay focused on security findings.

Where to see your usage

View live usage, included credits, overage and per-PR history at app.foundationmachines.ai/usage.

Spend cap and credit packs

Credits are bought in $25 / $100 / $500 packs. Bigger packs include bonus credits (up to +20% on the $500 pack). Purchased credits do not expire.

Team tenants set a spend cap. Sebastion emails warnings at 50%, 75%, 90% and 100% of your cap. At 100%, AI-backed work hard-stops: PR reviews, chat replies and autofix runs are paused before more credits are consumed.

If you want work to continue past the cap, open the billing portal, review the current usage and click the 24-hour override action. The override lets new usage continue and credits are consumed at the normal $0.02-per-credit base rate. You can disable the override (or lower the cap) any time from the same portal.