Boundless Opens Tempo Wallet Payments for AI Inference
On November 6, 2026, Boundless became the first inference provider on Tempo to accept machine-payments protocol (MPP) directly, allowing agents to purchase model responses using wallet assets without an account or API key.
The new service accepts three payment types: Tempo's native $MACH onchain credit, and the stablecoins $USDC.e and $OUSD. All assets are counted at dollar parity for billing purposes. When a wallet opens a session, it must make a deposit, and the final charge reflects only the tokens actually consumed by each request.
This implementation follows the machine-payments standard (MPP) co-authored by Stripe and Tempo. When an unpaid inference request is made, the system returns an HTTP 402 Payment Required response containing a payment challenge. The client must then open a payment channel and sign vouchers for each request.
Payment Process
Before running inference, Boundless reserves the worst-case cost of the request, counting every byte of the request body as an input token and adding the output limit. The final bill uses only the tokens the model actually consumed, with cached input charged at a reduced rate. Requests that the model fails to answer are fully refunded.
The default deposit is one token, with $USDC.e as the default currency on Tempo mainnet. Wallets funded with $MACH also need a small $USDC.e balance to cover network fees, which Boundless notes are minimal at a fraction of a cent. The provider recommends keeping deposits small since the endpoint moves real money on mainnet.
Client Requirements
Boundless currently supports only Nanocode as a client, which is an open-source agent SDK developed by Paradigm's Georgios Konstantopoulos. The prebuilt release doesn't yet include the Tempo gateway option, requiring users to build from source with a Rust toolchain. The client must implement the payment exchange process to work with the service.
Rate Structure
Boundless advertises sixteen open models over MPP at the same per-token prices as its existing API service. For example, DeepSeek V4 Flash costs $0.04 per million input tokens, $0.02 for cached input, and $0.21 for output. The cached rate applies when a request reuses a prefix already held by the gateway.
Unlike the existing team plan that uses prepaid credits, wallet-funded sessions let agents continue paying for inference without creating a Boundless account. Concurrent agents can share a wallet while maintaining separate payment channels, with each run tracked individually rather than against a shared API key.
Related
