What it is. fleeex is a standalone API that sits between an app and the AI models it calls. The app integrates it once, and from then on its end users pay for their own AI usage from a single prepaid wallet that works across every app built on fleeex. Behind each call, fleeex meters the exact cost, debits the wallet, and keeps a margin, so the app never has to build or run AI billing of its own. Arthur Robine and I are building it together; it's in active development.

• One global wallet per user. A person's prepaid balance (in EUR) is shared across every app they use through fleeex, and an app can only draw from a wallet that has explicitly authorized it.
• App-key auth. An app registers once for a server-side API key (shown once, stored hashed) and vouches for its own user on each call. fleeex trusts the app for its user's identity, never for wallet access.
• Metered LLM proxy. Calls go through an OpenAI-compatible endpoint, so an app can point the official OpenAI SDK straight at it. fleeex forwards to AWS Bedrock, measures the tokens against a versioned price table, and debits the wallet with a conditional write so it can't go negative under concurrent calls.
• Top-ups via Stripe. Users add credit through Stripe Checkout; the wallet is credited on the verified webhook, idempotently, never on the client redirect.

A containerized NestJS server on AWS ECS, with DynamoDB for the append-only usage ledger and the wallet, and AWS Bedrock behind the model endpoint. Rollups and reconciliation run off the hot path on EventBridge → SQS → Lambda. The stack is provisioned with Terraform, and the API ships an OpenAPI spec generated from its NestJS decorators.

We're launching fleeex as a two-person team. For now I'm on the tech side as CTO and tech lead (backend, architecture and product), and Arthur Robine handles DevOps and infrastructure.

NestJS (TypeScript), AWS ECS, DynamoDB, AWS Bedrock, Stripe, EventBridge / SQS / Lambda, Terraform, OpenAPI.