Urgent.News

What's breaking now, across thousands of outlets.

Editions

AI

Schema Gate: a signed acceptance boundary for AI-generated JSON

AI agents increasingly hand model-generated JSON directly to APIs, queues, databases, and other agents. Syntactically valid JSON is not enough: a missing field, an unexpected value, or a payload that violates a workflow rule can still trigger the wrong production action. I built Schema Gate to put a deterministic acceptance boundary between candidate machine output and the system that consumes…

AI agents are now sending model-generated JSON directly to various components such as APIs, queues, databases, and other agents. However, simply having syntactically valid JSON is not sufficient, as issues like missing fields, unexpected values, or payload violations that violate workflow rules can still cause incorrect production actions.

To address this, Schema Gate was developed to establish a deterministic acceptance boundary between the AI-generated JSON output and the system that consumes it. Schema Gate accepts candidate JSON, a bounded target schema, explicit acceptance criteria, a buyer-generated order ID, an idempotency key, and a commitment hash that ties the request to those rules.

When a valid unpaid request is made, Schema Gate returns a standard x402 v2 payment challenge. Upon settlement, it provides a machine-readable ACCEPT or REJECT decision with a signed ES256 receipt that ties the decision to the order, rules, input, output, price, and settlement. This receipt can be used by downstream workflows to ensure the decision is enforced before proceeding.

Unlike local schema validators, Schema Gate offers an independently signed decision that can be verified by another system. The launch price for Schema Gate is 0.010 USDC per completed signed evaluation on Solana mainnet through x402 v2, and both ACCEPT and REJECT outcomes are charged as the delivered product is the signed evaluation.

The client SDK is publicly available, keeping wallet signing local, and the repository includes a strict five-field acceptance policy, receipt verification, recovery handling, and examples. Schema Gate is currently live in managed cloud infrastructure with production and D1-backed acceptance checks in place. The service is seeking an independent workflow team to run a funded production evaluation and provide feedback on the integration's usefulness or any challenges faced.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in AI

Don't Let Codex Roam Free: 6 Guardrails I Use for AI-Assisted Coding

Leaving Everything to AI Feels Risky. But Codex Became Much More Reliable Once I Put Up Some Guardrails https://github.com/tosane932/sales_data_app Introduction This article was originally published…

  • Define Codex's scope to specific tasks like input validation and rollback handling
  • Write failing tests before modifying production code to reproduce and verify problems
  • Use isolated databases for testing to limit impact if issues arise during testing

DJI names winning developers in enterprise drone AI challenge

Technology manufacturer DJI announced the winners of its enterprise drone onboard artificial intelligence challenge on 20 August 2026 selecting solutions across industrial sectors

  • DJI announces 2026 Enterprise Drone Onboard AI Challenge winners.
  • Five Best Onboard AI Model winners and ten Industry Application Excellence winners.
  • Solutions applied to agriculture, public safety, and traffic management industries.

More from Friday 21 August →