Hi Product Hunt 👋 I'm Jeremy, solo founder. Built SteelSpine because my AI agent failed in production. I had logs. I still didn't know why. What it does: wrap any agent command (zero code changes) and capture every event into a hash-chained, Ed25519-signed audit trail. From that capture you can: 🔁 Replay step-by-step from any captured state 🔀 Compare two runs (line-by-line divergence highlighted) 🌳 Branch alternative futures from any event 🔐 Verify the chain wasn't tampered with Works with LangChain, LlamaIndex, MCP, raw Python, custom harnesses, anything you can wrap in a shell command. The captured chain doubles as EU AI Act Article 12-ready audit trails (Aug 2, 2026 enforcement) but the daily-use value is the debug loop. Local-first by design. Nothing leaves your machine. Works with any agent harness (LangChain, LlamaIndex, MCP, raw Python) and whichever model your agent calls. SteelSpine wraps your agent's process, not the API. The model is whatever your code talks to. CA$29.99/mo, 14-day free trial. Just me building this. Demo: https://asciinema.org/a/Ah79vmcO... Site: https://steelspine.ai HF technical writeup: https://huggingface.co/blog/Spec... PH discount: code PRODUCTHUNT for 25% off first 3 months. Happy to answer questions. Genuinely curious what PH builders use for debugging their AI agents today.

SteelSpine

Local-first replay tool for debugging AI agents

Local-first replay tool for debugging AI agents