DataGrout Invariant - Semantic code analysis for the AI era

Hey Product Hunt 👋 We've been building DataGrout for a while now, an intelligence layer that helps AI agents connect to enterprise systems, plan workflows, and execute safely. One thing kept coming up as we built it: AI agents write code, but they have no real way to verify whether what they wrote actually does what they intended. That's the problem Invariant solves. What it does: Invariant is a semantic code analysis tool built for AI agents (and the developers who build them). It's not just a linter. It uses a neuro-symbolic engine , combining LLM-based semantic understanding with deterministic Prolog rules, to give you answers you can actually trust. Here's what it can do in a single workflow: code_lens — analyze source code and extract structured facts: which functions call what, what their intent is, what side effects they have code_query — query those facts with a Prolog engine. No LLM at query time. Results are reproducible and instant. Find orphan functions, test gaps, dependency cycles, security concerns, intent mismatches diff_analyzer — give it code before and after a change plus the goal you stated, and it returns an alignment score telling you whether the change actually did what you said it would review — a full automated code review: criteria, constraints, pass/fail verdict, per-item reasoning The part we're most excited about is the agent feedback loop. You can drop two lines into an agent's system prompt and it will call diff_analyzer automatically after making changes. If the alignment score is below threshold or there are unexpected changes, the agent revises before presenting its work. It's a self-correction loop that doesn't rely on the agent trusting itself. Why we built this: We kept seeing agents produce code that was plausible but wrong in subtle ways. The goal drifted. Functions were added that weren't needed. Constraints were silently violated. You'd only catch it in review or worse, in production. Invariant makes goal alignment a checkable property, not a hope. There's also a CLI for local use in CI pipelines. Facts extracted locally get uploaded to DataGrout for server-side enrichment and querying. If you're running AI in your code review pipeline, tell me where it's breaking down, that's the kind of feedback that actually shapes what we build next. —The DataGrout team