Skip to Main Content

Precedent is a governance harness that lets company AI agents act, learn and be stopped, safely. Agents in companies fail in two ways: they do things they shouldn’t (install a web-search MCP that reads prompt-injected pages, query restricted data) and they forget what people taught them (every agent re-guesses the same undocumented warehouse table). In Precedent, the model only proposes; a deterministic policy engine decides every tool and MCP call before anything runs. Web-search MCPs are blocked outright (Claude judges, policy decides). Other MCPs become InfoSec requests ranked by how many people asked, with Raindrop evidence and a recommendation. When a user corrects an agent, a policy checkpoint runs and a human approves saving that lesson to MongoDB Atlas for their department, so the next engineer’s agent gets it right on the first try while other departments never see it. Precedent also drafts its own guardrails from repeated human approvals, replays them against recorded history, and has Raindrop verify the runs behind them before they ship. One click quarantines a compromised connector for every agent. The demo runs the same prompt side by side, without Precedent and with it. https://drive.google.com/drive/folders/1XlgovvrrhaZ9RycRNUXshxOfaGezh8Ch?usp=sharing

Team

Vikhassudhan30

More from The Harness Engineering & Model Wrangling #SFTechWeek Hackathon

MongoMation is an autonomous PLC engineer. You describe a machine in plain English, and Claude agents write IEC 61131-3 PLC code, compile and deploy it to a…

MongoMation project preview

LabRat is a self-improving harness for a physical lab agent, running on a real bench: two SO-101 arms, a 3018 CNC carrying a servo pipette I designed during…

LabRat project preview

Recursive Harness Scientist: an AI scientist that improves its own scientific method, and must replicate every gain on data it has never seen. Why it…

Recursive Harness Scientist project preview

Super Bid is a self-improving AI estimator that teaches itself to price bids the way a specific company would. It practices on the company’s real past bids…

SuperBid project preview

A robot that gets better at handling objects it’s never seen, without retraining. After every round, an AI coach reviews what broke and rewrites the robot’s…

WORLDSTATE gives physical spaces a memory. Any phone becomes an observer: open a link, point the camera, and WORLDSTATE tracks what's inside fridges, shelves…

WorldSTATE project preview