Comad World – 6개의 AI 에이전트가 하나의 YAML 구성에서 지식 그래프를 구축합니다.

Personal knowledge system that crawls RSS, papers & GitHub — then builds a searchable knowledge graph, updated daily. 6 AI agents that crawl → understand → simulate → curate → remember → automate for any domain you care about. Change one YAML file, get a whole new knowledge system. Quickstart · Architecture · Modules · Customization · Presets | Without Comad World | With Comad World | | |---|---|---| | Collecting | Manually check 20+ sites, forget half | ear auto-detects and archives from RSS, HN, arXiv, GitHub | | Organizing | Bookmarks pile up, no connections | brain builds a knowledge graph — 10,000+ nodes, searchable via GraphRAG | | Analyzing | Read each article, form opinions alone | eye runs simulations through 5 core strategic lenses (tiered system), tracks prediction accuracy | | Remembering | Context lost between sessions | sleep consolidates memory, voice automates recurring workflows | Key numbers from a real deployment - 60,000+ graph nodes, 150,000+ relationships from ongoing crawling - 22 RSS feeds monitored (OpenAI, Anthropic, Google, Meta, arXiv, researcher blogs) - 20+ MCP tools for querying your knowledge graph from Claude Code - Entity-level confidence scoring (0.0–1.0) for trust boundary tracking - Built-in performance monitoring via comad_brain_perf MCP tool - $0.60/day operating cost (down from $4.50 after optimization) - 1,482 tests across all modules Comad World is a modular AI agent system built on Claude Code. It connects six specialized agents into a pipeline that collects information, builds a knowledge graph, runs simulations, curates content, manages memory, and automates workflows — all driven by a single configuration file. ear (listen) → brain (think) → eye (predict) ↑ photo (edit) sleep (remember) voice (automate) The key idea: every domain-specific setting lives in comad.config.yaml . Swap the config, and the entire system adapts — from what RSS feeds to crawl, to what arXiv categories to watch, to how articles are classified. - Claude Code (Claude Max subscription recommended) - Docker (for Neo4j) - Bun (for brain module) - Python 3.13+ (for eye module) git clone https://github.com/kinkos1234/comad-world.git cd comad-world cp presets/ai-ml.yaml comad.config.yaml # or: web-dev, finance, biotech ./install.sh Then start collecting knowledge: cd brain && docker compose up -d && bun install && bun run setup bun run crawl:hn && bun run crawl:ingest # crawl & ingest bun run mcp # start MCP server # Start with AI/ML preset $ head -5 comad.config.yaml profile: name: "Comad AI Lab" language: "en" description: "AI/ML knowledge system" # Crawl AI sources (22 RSS feeds, 10 arXiv categories) $ cd brain && bun run crawl:hn [hn-crawler] Keywords: 48, RSS feeds: 22, HN queries: 8 [hn-crawler] HN stories: 347 [hn-crawler] RSS results: 412 [hn-crawler] Wrote 583 articles to data/articles-crawl.json # Now switch to Finance $ cp presets/finance.yaml comad.config.yaml $ ./scripts/apply-config.sh ✓ ear/interests.md ✓ ear/CLAUDE.md # Same crawl command, completely different sources $ bun run crawl:hn [hn-crawler] Keywords: 31, RSS feeds: 10, HN queries: 7 [hn-crawler] HN stories: 89 [hn-crawler] RSS results: 156 [hn-crawler] Wrote 201 articles to data/articles-crawl.json # ear/interests.md automatically updated: $ head -6 ear/interests.md # User Interest Profile ## High Priority (Core Focus) - Quantitative Finance (QuantConnect, Zipline, Backtrader) - Market Data / Analysis - DeFi / Crypto - Risk Management One YAML change. Different feeds, different keywords, different categories, different relevance criteria. ┌─────────────────────────────────────────────────────┐ │ comad.config.yaml │ │ (interests, sources, keywords, categories, stack) │ └───────────┬───────────┬───────────┬─────────────────┘ │ │ │ ┌───────▼──┐ ┌─────▼────┐ ┌──▼──────┐ │ ear │ │ brain │ │ eye │ │ (curate) │→ │ (graph) │→ │(predict)│ └──────────┘ └──────────┘ └─────────┘ │ ┌──────────┐ ┌────▼─────┐ ┌─────────┐ │ photo │ │ sleep │ │ voice │ │ (edit) │ │(remember)│ │(automate│ └──────────┘ └──────────┘ └─────────┘ - Ear detects articles in Discord, classifies relevance using your interests, archives to markdown - Brain crawls RSS/arXiv/GitHub filtered by your keywords, builds a Neo4j knowledge graph with entities and relationships. JS-heavy pages automatically rendered via Browse - Eye takes any text, converts to ontology, runs multi-round simulations, generates analysis through 5 core strategic lenses with prediction tracking - Photo corrects images via Photoshop MCP (domain-agnostic) - Sleep consolidates Claude Code memory across all projects (domain-agnostic) - Voice provides workflow automation triggers for Claude Code (domain-agnostic) - Search discovers GitHub repos, evaluates them, generates adoption plans, tests in sandbox — the system improves itself | Module | Config-Driven | Domain-Agnostic | |---|---|---| | ear | interests, categories, must-read stack, relevance thresholds | archive format, Discord integration, dige