mongoARC gives a Claude Code agent a long-term memory in MongoDB Atlas while it plays ARC-AGI-3, a benchmark of unfamiliar interactive games played level by level. Agents lose what they worked out when their context is compacted. Here the agent writes each lesson into Atlas in its own words, and searches that memory in plain language before it acts on the next level. Retrieval is MongoDB's own: Atlas Vector Search with automated embedding (Voyage) embeds every note and every question, so the harness sends only raw text. Notes are updated or deleted as later levels correct them, and memory is scoped to a single game, because ARC-AGI-3 knowledge transfers between levels, not between games. A live dashboard, read entirely from Atlas, replays every move frame by frame, shows which button or pixel the agent chose, its reasoning and code, and every memory write and search as it happens (via change streams).
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…

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…

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…

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…

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…
