Six finalist projects from the inaugural JetBrains x Codex Hackathon are showcased, each exploring AI agents natively embedded in the IDE. First place went to hyperreasoning, which replaces single-shot LLM calls with a visible search-tree approach using a learned controller and test feedback. Second place Scopecreep integrates hardware bring-up workflows into a single JetBrains tool window with human-in-the-loop probe placement. Third place mesh-code enables shared agent memory across machines. Other finalists include Periscope (context window management for long agent sessions), SecureLoop (automated security incident response with human approval gates), and Pinpoint (precise frontend feedback via on-page annotations). The common theme: Codex inside the IDE transforms it from a code editor into a visible, controllable reasoning layer.
Table of contents
🥇 First Place: hyperreasoning – Aditya Mangalampalli🥈 Second Place: Scopecreep – Bhavik Sheoran, Kenneth Ross, Roman Javadyan, Joon Im🥉 Third Place: mesh-code – Ayush Ojha, Coco Cao, Kush Ise, AL DRAMLatent Signal – PeriscopeSecureLoop – Abhiram Sribhashyam, Rahul Marri, Peyton LiPinpoint – Het PatelWhat the finalists showSort: