Reverse-engineering ChatGPT's memory: 4 layers, no vector DB, no RAG
HowDevelop · x · 2026-09-02
Developer Manthan Gupta reverse-engineered ChatGPT's memory system through conversation experiments and published it as the first of a 7-part blog series on agentic memory, covering ChatGPT, Claude, Hermes, OpenClaw, and voice-agent memory designs.
Key finding: ChatGPT's memory is far simpler than expected — no vector databases, no RAG over conversation history. Its context has six slots: system instructions, developer instructions, ephemeral session metadata, long-term user memory, summaries of recent chats, and the current session window.
Memory splits into four layers: ephemeral session metadata (device, browser, location, subscription tier), explicit long-term facts (the author found 33 stored facts about himself), lightweight summaries of past conversations, and a sliding window of the current chat. He argues this layered direct-injection approach may beat traditional retrieval systems. All findings come from behavioral reverse-engineering; OpenAI hasn't published these details.
More from coding & agent
- Agentic Engineering Is Constraint Engineering: The Case for a 'Constraint Engineer' Role — burny_tech · 2026-09-02
- Day-one review: Fable 5.1 a big step up for serious codebases, coordination shines — xeophon · 2026-09-02
- Codex draws on tldraw canvas via web-mcp in WebMCP Hackathon entry — HankYeomans · 2026-09-02
- WebMCP + agent-browser: agent-native web with tool discovery and browser fallback — HankYeomans · 2026-09-02
- Updating a skill is editing a file; updating agent memory is the real engineering problem — _jaydeepkarale · 2026-09-02
- Best vibe coding books for 2026: from quick prototypes to real software — Mahmoud_Zalt · 2026-09-02