Developer builds a persistent AI memory prototype in regular ChatGPT chat, state stored on Google Drive
epicskyes · reddit · 2026-09-15
A developer describes a persistent-memory prototype built through the regular ChatGPT chat interface (not Work mode), with Google Drive holding the saved project state. The conversation itself built the backend, wired components, ran tests, and read back results.
The key design: memory lives outside the model. Project state is stored externally and retrieved on demand rather than relying on the model's internal memory. Records are structured and connected — what information exists, where it came from, how things relate, and what changed — with original sources kept separate so working memory stays traceable instead of degrading into disconnected summaries. Work runs during sessions; saved memory persists in Drive. It is not an always-on background agent.
Still a prototype with implementation details kept private, but the milestone is a working persistent-memory backend actually built and tested from the ordinary chat interface.
More from coding & agent
- GPT-6 Astra says you can delete scaffolding — just not the control kind — Slight_Republic_4242 · 2026-09-15
- You Don't Need Ultra: Power User's Rule of Thumb Says Low/Medium Tiers Suffice for Daily Tasks — soumitrashukla9 · 2026-09-15
- Stop Asking AI Agents 'Can You Build This?' — Use This Inspection-First Prompt Instead — taherdhanera · 2026-09-15
- Claude Cowork 101: A Beginner Tutorial Written for Academic Researchers — JeremyNguyenPhD · 2026-09-15
- ASC: a fast Android decompiler front-end built for AI agents, 1K stars in days — MG1937 · 2026-09-15
- AI agent toolkit pi hits 105K GitHub stars with unified LLM API and coding agent CLI — earendil-works · 2026-09-15