666ghj/mirofish
A Simple and Universal Swarm Intelligence Engine, Predicting Anything. 简洁通用的群体智能引擎，预测万物
How mirofish is put together
MiroFish is a multi-agent AI 'swarm intelligence' prediction engine that ingests seed materials (PDF/MD/TXT), builds a knowledge graph, spins up thousands of LLM-driven social agents to simulate parallel digital worlds, and generates prediction reports. It is a two-tier app: a Vue 3 (Vite) frontend and a Python Flask backend organized as an application-factory with blueprints for graph building, simulation, and reporting. The backend orchestrates LLM calls via the OpenAI SDK, persists agent memory in Zep Cloud (GraphRAG), and runs social simulations on the OASIS/CAMEL-AI framework in managed subprocesses. Data flows: upload → ontology generation → graph build (Zep) → agent/persona config → OASIS simulation → ReportAgent interaction. Frontend proxies /api to the Flask backend on port 5001.
Languages
Frameworks
Datastores
Infrastructure
Major components
Flask API layer (backend/app/api)
Exposes graph, simulation, and report blueprints under /api/* handling uploads, project lifecycle, and simulation control.
Graph building service
Generates ontologies from uploaded documents and constructs GraphRAG memory graphs in Zep Cloud.
Simulation engine (services)
Runs OASIS-based multi-agent social simulations in managed subprocesses with IPC, config, and profile generation.
ReportAgent service
LLM agent with a toolset that interacts with the post-simulation environment to produce prediction reports.
Zep integration layer
Manages Zep Cloud graph lifecycle, entity reading, memory updates, tools, and paging.
Models (project/task)
Defines and persists Project and Task domain state with managers and status enums.
Vue frontend
Multi-step workflow UI (graph build → env setup → simulation → report → interaction) with graph visualization and i18n.
Utilities (utils)
Cross-cutting helpers: file parsing, LLM client, logging, retry, locale, and OpenAI chat compatibility.
A routine automated update to the project's star-count chart.
5 weeksWeek of 2026-08-17
Week of 2026-08-17latest
A routine automated update to the project's star-count chart.
Week of 2026-08-03
Made the automated star-history updates run safely on the protected main branch.
Week of 2026-07-20
A large wave of reliability and security work, especially around the external knowledge-graph (Zep) integration and simulation data handling.
Week of 2026-04-27
Added support for building the app's Docker image on ARM64 hardware.
Week of 2026-04-20
Two targeted fixes to prevent crashes and AI hallucinations.
The most recent weeks have been dominated by automated 'Star History' bookkeeping — a small chart tracking the project's GitHub popularity — with fixes to make those updates run safely without a human. The one big burst of real engineering was in mid-July, when the team hardened the app's integration with an external knowledge-graph service (Zep), fixed several data-handling and simulation bugs, and closed out security vulnerabilities. Earlier spring work laid groundwork with Docker builds and a couple of crash fixes.
Week by week
2026-08-17A routine automated update to the project's star-count chart.latest1 change
Chore
Star history refreshed
The chart tracking the project's GitHub stars was updated automatically, recording a jump to roughly 71,000 stars.
2026-08-03Made the automated star-history updates run safely on the protected main branch.3 changes
Fix
Star updates now go through pull requests
Automated changes to the star chart are submitted as reviewable pull requests instead of pushed directly, so they respect branch protections.
Fix
Fixed star updates on protected main
Resolved an issue that prevented the automated star-history job from working on the safeguarded main branch.
Chore
Star history refreshed
The star-count chart was updated with the latest figures.
2026-07-20A large wave of reliability and security work, especially around the external knowledge-graph (Zep) integration and simulation data handling.6 changes
Fix
Modernized the Zep knowledge-graph integration
Overhauled how the app talks to the external Zep service, using supported methods, safe pagination, and internal timeouts for more reliable results.
Fix
Cleaned up profile and ontology data
Generated profiles and knowledge-graph fields are now normalized consistently so downstream services receive well-formed data.
Fix
Closed a security vulnerability
Patched CVE-2025-14009 and updated vulnerable JavaScript dependencies to keep the project secure.
Fix
Guarded against fabricated AI results
The system now strips made-up tool results everywhere to prevent the AI from acting on hallucinated information.
Fix
More robust simulation handling
Simulations now stop cleanly after preparation failures, prioritize failure status correctly, and can load Twitter-only profiles from a CSV file.
Chore
Safer Docker publishing and dependency upgrades
Made ARM64 Docker image publishing opt-in and upgraded the NLTK language toolkit, alongside disabling Flask debug mode by default.
2026-04-27Added support for building the app's Docker image on ARM64 hardware.1 change
Chore
ARM64 Docker builds
The project can now produce Docker images for ARM64 processors, broadening where it can run.
2026-04-20Two targeted fixes to prevent crashes and AI hallucinations.2 changes
Fix
Prevented a knowledge-graph crash
The app now handles text attributes in the graph ontology correctly instead of crashing with a type error.
Fix
Removed fabricated AI results
Made-up tool-result blocks are stripped out to stop the AI model from hallucinating.
Dependencies and code review
Dependency advisories
Security Watch
No known advisories across 25 scanned dependencies.
No known advisories in the scanned dependencies.
Code review
No concrete code-level issues confirmed.
Get this report every week for your repos.
GitZoid learns each repo, reports what changed, and flags what needs attention. One flat price for the whole team.
$19 a month, flat · First 10 outputs free · No card required