ChatGPT silently forgets the start of long threads; dev fixes it with external memory over MCP
Asly97 · reddit · 2026-09-19
A developer two weeks into a backend migration found that endless ChatGPT threads quietly lose earlier decisions—the model confidently claimed a rejected database had been "decided," as if the first fifty messages never existed.
He tried pasting daily summaries into fresh chats and the DECISIONS.md trick; both failed as summaries bloated and the model promised updates but never wrote them.
What finally worked: storing decisions and their rationale in an external memory tool (Vilix AI) that all his AI tools read over MCP, so decisions no longer live inside any single context window. Caveats: models are lazy about saving (he often has to say "save that"), and setup took 40 minutes after fumbling OAuth twice.
More from coding & agent
- Human Oversight Is 70-75% of AI Agent Costs in Banking, McKinsey Finds — mikeflache · 2026-11-04
- Over 80% of Major Financial Institutions Now Run Agentic AI in Production — mikeflache · 2026-11-04
- Mnemos MCP nears launch after shared memory layer fix — RileyRalmuto · 2026-11-04
- Moving mobile dev to remote machines makes it multiplayer across product, design, engineering — rudrank · 2026-11-04
- Kuaishou's EvoRec: self-evolving agentic recommender lifts ad revenue 4.6% — _reachsumit · 2026-11-04
- TRIO: three specialized agents with aligned objectives refine retrieval-augmented generation — _reachsumit · 2026-11-04