FIND-20260323-023 · 2026-03-23 · Innovation Veille

MiroFish — Multi-Agent Swarm Intelligence Prediction Engine

adhoc MEDIUM
MiroFish is an open-source multi-agent simulation platform that builds digital worlds from seed data (news, reports, documents) to predict future outcomes. Users describe a scenario in natural language; the engine spawns thousands of autonomous AI agents with unique personas and long-term memory, runs a parallel dual-platform simulation, and produces structured prediction reports plus interactive post-run chat with individual agents. The backend is Python, the frontend is Vue.js, and it is built on the OASIS framework with Zep Memory Graph for agent memory persistence. It natively supports OpenAI-compatible APIs and Claude Code headless mode as the LLM backend. The repo is a fork (63 stars) of a much larger upstream project (40k+ stars, 5.5k forks), created March 13 2026 and last updated March 20 2026. License is AGPL-3.0, which requires legal review before any integration into ODS Platform.

Source

https://github.com/JayFarei/MiroFish

ODS Impact

Not directly applicable to ODS core microservices stack (Rust/Go backend, Redpanda, PostgreSQL). However, the multi-agent simulation pattern is architecturally interesting for two future ODS scenarios: (1) the Workflow Engine could adopt agent-based parallel task execution patterns; (2) a future AI/analytics layer could leverage swarm intelligence for tenant behavioral prediction or document processing scenario testing. The AGPL-3.0 license is a hard blocker for direct embedding — any use must be via isolated API calls to a separately hosted instance. The Python-only stack does not align with ODS core service technology choices.

Security Review

License: AGPL-3.0 | Maintenance: ACTIVE | Risk: HIGH | Recommendation: DO_NOT_USE

Tags

ai multi-agent swarm-intelligence python llm simulation prediction agpl