Aura

Semantic version control for AI coding agents on top of Git

Developer Tools

Artificial Intelligence

GitHub

Vibe coding

Hey Product Hunt! 👋 I'm Mo, CEO of Naridon (naridon.com), and today we’re open-sourcing Aura. At Naridon, our main business is building complex AI Search Optimization (AIO) infrastructure for e-commerce brands. We spend our days working deeply with LLMs to optimize how models like ChatGPT and Perplexity index and recommend products. Because we build AI products, we rely heavily on autonomous AI agents (like Cursor and Claude) to write our code. But over the last year, we hit a massive bottleneck: Git was built for humans typing linearly, not for AI agents generating 4,000 lines of non-linear code per minute. When our AI agents hallucinated, standard text diffs resulted in chaotic, unresolvable merge conflicts that brought our sprints to a halt. We had to build Aura for our own team's sanity. It is a "Semantic Time Machine" that stops AI agents from breaking our production environments. Today, we’re sharing it with the world for the betterment of the agentic coding future. Instead of tracking text lines, Aura natively parses your codebase into an Abstract Syntax Tree (AST) locally (supporting Rust, Python, TypeScript, and JavaScript). 🚀 What Aura gives you for free (Apache 2.0): * The Semantic Scalpel (`aura rewind`): Revert a single broken function or class the AI wrote without losing the rest of the good code in the file. * The Amnesia Protocol (`--amnesia`): Surgically wipe an AI's chat memory of a specific coding hallucination so it doesn't get stuck in a recursive failure loop. * The Gatekeeper (`aura capture-context`): A parasitic Git hook that hard-blocks git commit if the AI's natural language intent doesn't mathematically match the AST nodes it modified. * Native GSD Orchestration (`aura plan`): We integrated the "Get Shit Done" methodology directly into the Rust core. It X-Rays your AST Merkle-Graph and builds mathematically sound execution waves before the AI writes a single line of code. * The Sovereign Allowlist (`aura request-access`): Securely whitelist specific logic nodes (like Auth Headers) to bypass the Gatekeeper, allowing for precise secrets management. * Semantic Audit (`aura audit`): Scans your Git history to catch any rogue, undocumented code an AI agent snuck in using --no-verify. * Token Efficiency (`aura handover`): Compresses your entire architectural context into dense XML, saving you up to 95% on LLM API token costs when switching agents. Aura operates as a meta-layer directly on top of Git. It runs 100% locally on your machine, we never see your code. We’ve released the core engine today under the Apache 2.0 license. This isn't our core commercial product; it's the foundational tool we had to build to survive the AI era, and we wanted the community to have it. Would love your feedback! Try it out with a single curl command on macOS/Linux: curl -fsSL https://auravcs.com/install.sh | bash Question for the community: What's the worst merge conflict an AI agent has caused you recently? Let me know below! 👇