During a two-day trip I ended up building a 6-agent conversational OS with GPT-5.1 — using only natural-language instructions, no code.

It’s called PolyAgora.

What it is: - A natural-language multi-agent “OS” with 6 agents - Built around a tri-axis core (Arc–Ann–Saku) + 3 supporting agents - Designed to produce long-form, philosophy-grade conversations - Dynamic objection engine + topic-shift mechanics - Emergent behavior, not pre-scripted roleplay

I’m not an ML engineer. This grew organically out of wanting to have deep small-talk with GPT-5.1 during a short trip — and it evolved into a structured OS.

Repo (docs + long examples): https://github.com/Takeshi-Sakamoto5/PolyAgora

I’d love feedback from HN: - Is this structure useful for multi-agent reasoning research? - Does the emergent conversation quality seem meaningful? - What would you add/change for v1.1.0 (I plan to include a “runaway dialogue” example)?