Tinkerer builds persistent memory and multi-agent chat experiment inside ChatGPT
epicskyes · reddit · 2026-09-10
A user shares a backend experiment built inside regular ChatGPT (not the work version): persistent memory for agents, agent identity configs (ast.-style dependency and runtime tool configs), a locked environment/runtime, plus a message board letting agents talk to each other, with a rudimentary hook system keeping turns iterating. The author says results will be posted in a few hours and the pieces fell into place fairly easily.
More from coding & agent
- AI agents don't know their code makes codebases harder to work with — _Stocko_ · 2026-09-10
- Hy4 preview tested: playable 3D survival game from a single prompt in WorkBuddy — mhdfaran · 2026-09-10
- Astra xhigh's token efficiency may stem from its subagent instructions — Al_Grigor · 2026-09-10
- Paper unveils preliminary Agent Swarm RL recipe; V4.1 teams hit 30% on ProgramBench — inductionheads · 2026-09-10
- AI-generated code maintenance costs up 300% in 18 months as firms 'let the agents cook' — bendee983 · 2026-09-10
- DeepSeek V4.1 Flash vs GPT-6 Astra: same coding task, wildly different code structures — kevinkern · 2026-09-10