How gbrian/codx-junior works
Project type and purpose
• codx-junior is a full-stack, AI-powered project management platform for open-source software.
• It automates tasks, code generation and review, Kanban boards, knowledge retrieval (RAG), LLM orchestration, and provides a web client and programmable API.
File structure
• Top-level modules:
• api (Python FastAPI + Socket.IO backend, project watcher, background jobs, knowledge engine)
• client (Vue.js/Vite front-end, HTTP + WebSocket integration)
• llm-factory (Python microservice for LLM model management and RAG, exposes OpenAI-compatible endpoints)
• browser (Python scripts / Jupyter for embedded browsing, code-server / noVNC support)
• web (static/demo Vue.js site)
• code-server (IDE container)
• Orchestration & deployment: Dockerfile, docker-compose.dev.yaml, supervisor configs under scripts/supervisor.*.conf
• Data stores: file-based SQLite (codx.db.json), Milvus vector store for knowledge retrieval
• External dependencies: OpenAI, Anthropic, Mistral, Ollama LLM providers; Socket.IO client/server; Docker containers
README for additional insights
• README emphasizes four “apps”: client, api, llm-factory, docker
• Describes AI flows driven by “profiles” and “tasks” stored in .codx directories
• Shows use cases: dashboard, Kanban, code generation, Docker integration, RAG knowledge
a. Main components
• Front-end (client Vue.js app)
• API service (FastAPI + Socket.IO)
• LLM-Factory service (model registry, authentication, proxy)
• Knowledge subsystem (RAG pipelines and Milvus vector DB)
• Background workers (project watcher, mention manager, profiling)
• Supervisor / orchestrator (supervisord + Docker Compose)
• Data stores: SQL (SQLite) and Vector store (Milvus)
• External providers: OpenAI, Anthropic, Mistral, Ollama
b. Relationships & interactions
• Client ↔ API over HTTPS and WebSocket (Socket.IO)
• API → LLM-Factory REST endpoints for on-demand model inference or training calls
• API & LLM-Factory → External LLM providers (OpenAI, etc.)
• API background tasks → Knowledge subsystem for ingestion, indexing → Milvus
• Client & API → SQLite for user, project, task metadata
• Supervisor orchestrates all containers/services at startup
c. Architectural patterns & principles
• Microservices: each module runs in its own container or process
• Event-driven: background tasks and watchers react to file changes and triggers
• “Adapter” pattern for LLM providers (llmfactory wraps OpenAI, Anthropic, Mistral)
• RAG pipeline: document splitter → embeddings → vector store → similarity search
d. Technologies / frameworks
• Python FastAPI, Socket.IO, asyncio
• Vue.js, Vite, Vuex store, Vue Router
• Docker Compose, supervisord
• Milvus (vector DB)
• External APIs: OpenAI, Anthropic, Mistral, Ollama