Product
AI credit
Permanent topup credit, legacy grants, billing, and rate limits.
A request in the ledger
Reserve
Input + maximum output
Settle
Usage at reserved prices
Product
Permanent topup credit, legacy grants, billing, and rate limits.
A request in the ledger
Reserve
Input + maximum output
Settle
Usage at reserved prices
Release
Unused credit returns
Every finalized $1 topup grants $1 AI credit. New topup credit never expires. Credit is tied to the paying wallet and cannot be withdrawn as cash or transferred through the app.
Chat and every API key in your wallet use one balance. A .ai Agent ID is required for AI access. Signing in proves wallet ownership but does not create a credit grant.
A key's lifetime budget is a ceiling on that key's spending, not an additional balance. Adding wallet credit does not reset its budget. See API keys.
Admins configure model IDs, availability, input/output prices per million tokens, and maximum output. Chat displays model prices. The API models endpoint returns current pricing.
Before a request starts, the ledger reserves enough credit for bounded input and maximum output. Requests already in flight retain their pricing snapshot. When usage arrives, the ledger charges actual usage up to the reservation and releases the rest. Provider failures before output release reserved credit. Interrupted streams can charge observed output. Abandoned reservations release after five minutes.
Prices use integer microdollars, with usage rounded up to the nearest microdollar. Concurrent requests cannot spend the same credit. The wallet rate limit is 30 requests per minute across Chat and API keys; keys may set lower limits or budgets.
Choose a model in the bottom composer. Its menu shows input/output pricing and the maximum output length. Auto uses the provider's defaults. Low, Medium, and High are available only when the model declares support for those effort levels. Changing models resets thinking to Auto.
More reasoning can take longer and use more credit. Reasoning tokens count toward the output limit and billable output even though Chat only displays the answer. If a successful Chat response omits valid output usage, billing uses the reserved maximum output, capped by the reservation.
Chat saves messages, titles, drafts, the selected model, and thinking level per wallet in localStorage. Open, rename, or delete a conversation from the sidebar. New chat starts a separate conversation. Sign-out and wallet changes hide the previous wallet's history; signing in again restores it. The Chat guide explains the controls and context limits.
History stays in the current browser and does not sync between devices. Clearing browser data removes it. If browser storage is blocked or full, Chat shows an error and new changes may only last for the current session. Prompts and recent conversation context are sent to the model provider.
An interrupted or failed message can be retried. Retry sends a new request. A request that was already accepted by the server can still finish and use credit after closing the page or deleting its local conversation.
The topup migration retains previous grants, recorded spending, and original expiry dates. Expiring legacy credit is consumed before new non-expiring topup credit. The dashboard shows the next legacy expiry when applicable.
A successful payment may be waiting for finality or the indexer. The settlement worker requires RPC finality and at least 15 blocks after payment. It validates the canonical receipt and grants each event once. Check Topup history before submitting another payment.