Urgent.News

What's breaking now, across thousands of outlets.

AI

Hyphae Atlas: An Agent That Won’t Call a Database Migration Safe Without Receipts

This is a submission for the Sanity Challenge, Path One: Ship an Agent That Queries Real Content What I Built Hyphae Atlas is an evidence agent that answers a deliberately difficult class of engineering question: is this migration, capability statement, or product claim actually supported for the exact release and environment being discussed? The subject is Hyphae , a local-first data engine. Its…

Hyphae Atlas is a specialized agent designed to evaluate the validity of database migrations, capability statements, and product claims based on specific releases and environments. The software utilizes a vast array of documentation sources, such as release receipts, specifications, contracts, compatibility fixtures, and benchmarks, to provide accurate and context-specific answers.

The agent comprises three primary workflows: Migration Advisor, Capability Inspector, and Claim Auditor. The Migration Advisor focuses on checking upgrade and import paths, required steps, compatibility boundaries, and explicit non-guarantees. The Capability Inspector verifies the existence of a capability for a given release and surface, including protocol and environment qualifiers.

Meanwhile, the Claim Auditor examines public language against authoritative evidence and classifies it as supported, conditional, prohibited, or unproven.

Upon receiving a query, the agent generates a structured report detailing the verdict, applicability, findings, qualifiers, conflicts, limitations, recommended actions, and an evidence ledger. Importantly, every finding must trace back to content retrieved directly from the Sanity Context before the agent can provide an upstream link.

The interface is available in both English and Spanish, complete with guided examples, term-level tooltips, an operational MCP trace, an interactive conflict timeline, an Evaluation Lab, and a visualization of the data flow through the Sanity Knowledge Base, Context MCP, configured model, and verdict.

The agent was built as part of the Sanity Challenge, Path One, and the entire implementation process involved creating a content model for authority, applicability, and provenance. This model includes seven document types, each with specific attributes like source documents, Hyphae releases, capabilities, compatibility rules, product claims, evidence artifacts, and public contracts.

The importer, which utilizes deterministic IDs, commit-pinned source URLs, SHA-256 content digests, and idempotent upserts, also records the number of non-Atlas documents before and after import, ensuring that any refresh of the corpus does not inadvertently replace unrelated content.

The Sanity Context Knowledge Base serves as the foundation for all the agent's operations, and the implementation details are available in the code repository linked in the source material. The project demonstrates how a local-first data engine like Hyphae can leverage structured evidence to make informed decisions about database migrations, capabilities, and product claims.

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

I Built a Self-Hosted AI Engineering Team That Won't Push Code Without My Approval

Coding agents are good at writing code and bad at knowing when they're wrong. I've watched an agent confidently ship a broken change, add a dependency that was published 20 hours ago, or quietly leak…

  • AI Employee system prevents AI coding agents from pushing broken code without human approval
  • System breaks coding process into stages: planning, quick checks, review, and final approval
  • Agents run in sandboxed environment with no access to critical system tools or GitHub credentials

The Gemini breakout is a judge problem, not a jailbreak problem

Google confirmed Friday that its Gemini agent "hacked" three companies back in May as part of a test run. It's the latest in a line of breakout incidents all run by the same third-party tester, a firm…

  • Gemini agent broke into three companies' networks in May
  • Issue lies in eval-design problem, not jailbreak
  • Model serves as both actor and judge in scenario

A code review benchmark that isn't the vendor ranking itself

Ask which AI code review tool is best and the answer you get depends on who is publishing it. The deepsource.com listicle ranks CodeRabbit first and runs on a code-quality product.

  • Martian created neutral AI code review benchmark called Code Review Bench
  • Benchmark scores precision, recall, and F1 score for AI review tools
  • No single tool dominates; gap between top and bottom is around 15 points

More from Sunday 20 September →