Architecture teardown of MiroFish source code

[NeuZhou]

I analyzed MiroFish's architecture from the source.

The main finding: the collective intelligence feature is LLMs role-playing humans on a simulated social network via OASIS multi-agent simulation. The report_agent.py at 1,400 lines is the largest single file.

Full writeup: https://github.com/NeuZhou/awesome-ai-anatomy/tree/main/mirofish

[666ghj]

Note

🤖 Automated maintainer response — Generated by the MiroFish triage agent and checked against main@4ebb2ed6 on 2026-07-21. Posted with maintainer authorization; a maintainer may revise it.

Thanks for sharing the source-code analysis. It is useful as an independent snapshot, but it should not be treated as official architecture documentation.

The file-size statement is now stale: on the reviewed current main, backend/app/services/report_agent.py is 2,572 lines, while backend/app/api/simulation.py is 2,716 lines. More generally, describing the system only as “LLMs role-playing humans” does not cover the graph and memory, profile and configuration generation, OASIS simulation, and ReportAgent layers.

Please pin the exact MiroFish commit analyzed and distinguish source facts from architectural interpretation. We are moving this Discussion to Show and tell and keeping it open.