# Build with AI

## Turn a conversation into a requirement

Describe the bot's audience, commands, and what should happen on unusual input. For example: “A private task bot with `/add`, `/list`, and `/done`. Each user only sees their own tasks.” AI may ask questions before proposing a summary.

Chat does not generate or deploy code. Review and edit the requirement card, then press **Generate a version**.

## Daily shared allowance

Free early access includes 20 AI interactions per account per UTC calendar day, shared across all bots. One requirement chat or one generation/change uses one interaction. The allowance resets at 00:00 UTC; the former monthly allowance is inactive. Failed chats with no output are refunded.

## What a generation includes

One generation includes the initial attempt and up to two automatic repairs. The quota is charged when execution begins. Resuming a generation task and internal automatic repairs do not consume extra interactions. Cancelling a queued task before it has ever started does not consume an interaction. Infrastructure failures are refunded; model or validation failures after execution begins count toward the daily shared allowance.

Deployments are unlimited during free early access, but safety checks, concurrency limits, and token budgets still apply. One generation can run per account at a time. Each task has a 10-minute deadline and a token budget. Chat and builds share the platform's daily model budget. If the service is not configured or a limit is reached, TeeAI reports that condition instead of generating a fake result.

## Supported code

New and changed modules use TypeScript. Untouched imported JavaScript is retained exactly. The validator uses a fixed SDK surface; arbitrary npm modules and Node/Bun runtime APIs are unavailable on Telegram Serverless.

The first version focuses on menus, simple community helpers, polls, and user-specific persistent tools. New paid runtime AI integrations are outside the beta. Source is available under the latest version or version history.

## Validation boundaries

Type checking, module checks, and isolated message/callback simulations run separately from the coding agent. Review the report. Test real command behavior, permissions, and database persistence in Telegram after publishing.