{
  "markdown": "# DiPeO, Diagrammed People (agents) & Organizations (agent system)\n\n* The whole codebase was created by vibe-coding, with Claude code & Codex.\n\n> Start with `dipeo ask --to \"the command you want to ask\" --and-run`\n\n![Watch Demo](./docs/pics/ad.webp)\n\n* This will generate diagram and run it as you want to create.\n\n### Why diagram?\n\n* To show how it is **structured**, and how it works **in realtime**.\n* For consistent outputs, rather than asking to code.\n\n### Hmm, so can I tweak it?\n\n<div style=\"text-align: center;\"><img src=\"/docs/pics/img.png\" width=\"85%\" alt=\"\"></div>\n\n\n* Definitely. You can just run `make dev-web` and tweak it in `localhost:3000`\n* The whole procedure works inside your computer. Nothing is in cloud or somewhere in network.\n\n### How about the diagram file?\n\n* You can see the diagram file in `.yaml`\n\n<details>\n    <summary>The diagram file content we created looks like...</summary>\n\n```yaml\nversion: light\nnodes:\n- label: start\n  type: start\n  position:\n    x: 330\n    y: 200\n  trigger_mode: manual\n- label: printer\n  type: person_job\n  position:\n    x: 632\n    y: 231\n  default_prompt: say hi\n  max_iteration: 3\n  memorize_to: ALL_MESSAGES\n  person: person 1\n- label: endpoint\n  type: endpoint\n  position:\n    x: 0\n    y: 437\n  file_format: txt\n  save_to_file: true\n  file_path: files/results/total.txt\nconnections:\n- from: start\n  to: printer\n  content_type: raw_text\n- from: printer\n  to: endpoint\n  content_type: raw_text\npersons:\n  person 1:\n    service: openai\n    model: gpt-5-nano-2025-08-07\n    api_key_id: APIKEY_52609F\n```\n\n</details>\n\n* For example, the diagram you generated with `dipeo ask ...` will be placed in `projects/dipeodipeo/generated`\n* We support `.light.yaml` format which will format the diagram in readable format.\n\n### How can I start?\n\n```bash\n# Clone the repository first\nmake install          # Install all dependencies (installs uv if needed)\nmake graphql-schema   # Generate GraphQL types\nmake dev-all          # Start both frontend and backend servers\n```\n\n### Ok. So is there a rule for diagram? Or, would you explain more detail?\n\n* Yes, here is the documentary in detail.\n  - [Full Documentation Index](docs/index.md) - Complete list of guides and technical documentation\n  - [User Guide](docs/README.md) - Getting started with DiPeO diagram editor\n  - We are developing some interesting projects using `dipeo` itself. Take a look at [projects](docs/projects)\n\n---\n\n> DiPeO(daɪpiːɔː) is a **monorepo** for building, executing, and monitoring AI‑powered agent workflows through an intuitive visual programming environment. The repository is composed of a feature-based React **frontend** (apps/web/), a domain-driven FastAPI **backend** (apps/server/), and a CLI **tool** (apps/server/src/dipeo_server/cli/) that work together to deliver real‑time, multi‑LLM automation at scale.\n\n## 핵심 기능\n\n1. LLM과 작업 블록의 분리를 통한 직관적인 컨텍스트 관리\n2. diagram의 yaml 형태 표현 및 실행 tool 제공\n3. 다이어그램 엔드포인트를 활용한 A2A canvas 제공 (구현 예정)\n\nFor motivations, guide, details in Korean, read [Korean docs](docs/index.md)\n\n### Code Generation (For Development)\n\nIf you need to modify the codebase or add new features:\n\n```bash\n# After modifying TypeScript specifications in /dipeo/models/src/\ncd dipeo/models && pnpm build     # Build TypeScript models\nmake codegen                      # Generate code (includes parse-typescript)\nmake diff-staged                  # Review changes\nmake apply-syntax-only            # Apply staged changes\nmake graphql-schema              # Update GraphQL types\n```\n\n## Major Features\n\n### 1. Claude Code Integration\n\nDiPeO features built-in support for Anthropic's Claude Code SDK, enabling seamless integration with Claude's advanced AI capabilities. Simply configure your diagram with Claude Code agents:\n\n```yaml\npersons:\n  Frontend Generator:\n    service: claude-code\n    model: claude-code\n    api_key_id: APIKEY_CLAUDE\n    system_prompt: |\n        You are an expert React/TypeScript engineer.\n        Generate clean, production-ready code.\n```\n\n**Key Benefits:**\n- Streaming-first architecture for real-time responses\n- Built-in conversation management\n- Automatic retry logic with exponential backoff\n- Context manager pattern for efficient resource usage\n\nFor detailed setup and usage, see [Claude Code Integration Guide](docs/integrations/claude-code.md).\n\n### 2. Frontend Auto - Rapid Application Generation\n\nGenerate complete, production-ready React applications in 30 minutes with Frontend Auto. This streamlined system creates fully deployable frontends with modern tech stack:\n\n```bash\n# Generate a complete chat application\ndipeo run projects/frontend_auto/consolidated_generator --light --debug --timeout=120\n\n# Generate with specific variant (e.g., e-commerce, analytics, banking)\ndipeo run projects/frontend_auto/consolidated_generator --light --debug --timeout=120 \\\n  --input-data '{\"config_file\": \"variants/ecommerce_config.json\"}'\n```\n\n**Generated Features:**\n- React 18 + TypeScript + Vite\n- Tailwind CSS styling\n- TanStack Query for data fetching\n- React Router v6 navigation\n- Complete component architecture (atoms/molecules/organisms)\n- Mock API with real-time features\n- Vercel-ready deployment configuration\n\n**Available Variants:** Chat applications, e-commerce stores, analytics dashboards, banking portals, CMS systems, healthcare portals, learning platforms, project management tools, and more.\n\nFor comprehensive details, see the Frontend Auto project files in `projects/frontend_auto/`.\n\n### 3. Multi-LLM Support\n\nBeyond Claude Code, DiPeO supports multiple LLM providers:\n\n**Ollama (Local Models):**\n```yaml\npersons:\n  Frontend Generator:\n    service: ollama\n    model: gpt-oss:20b\n    api_key_id: APIKEY_OLLAMA\n    system_prompt: |\n        You are an expert React/TypeScript engineer.\n        Generate clean, production-ready code.\n```\n\n**Custom APIs and Services:**\n- Notion integration\n- Custom LLM endpoints via cURL\n- Any RESTful API service\n\nThanks to our schema-driven integration system, adding new external API features is straightforward. See `integrations/` directory for examples.\n\nFor local model examples, see [Ollama example](examples/simple_diagrams/simple_iter_ollama.light.yaml).\n\n\n### `dipeo` - Run Diagrams with CLI\n\n#### Run existing diagrams\n```bash\n# Run diagram with automatic server startup\ndipeo run examples/simple_diagrams/simple_iter --light --debug --timeout=25\n# Run with specific diagram file\ndipeo run examples/simple_diagrams/simple_iter.light.yaml --light --debug\n# Run with input data\ndipeo run [diagram] --input-data '{\"key\": \"value\"}' --light --debug\n```\n\n#### Generate diagrams from natural language\n```bash\n# Generate a diagram from natural language request\ndipeo ask --to \"create csv preprocessor\" --timeout=90\n\n# Generate and immediately run the created diagram\ndipeo ask --to \"create csv preprocessor\" --and-run --timeout=90\n\n# With additional options\ndipeo ask --to \"build data pipeline\" --and-run --timeout=120\n```\n\n**Note**: The `dipeo ask` command uses AI to generate DiPeO diagrams from your natural language description. Generation typically takes 150-250 seconds due to multiple LLM calls.\n\n#### Convert Claude Code Sessions to Diagrams\n```bash\n# List your recent Claude Code sessions\ndipeocc list --limit 50\n\n# Convert the latest session to a DiPeO diagram\ndipeocc convert --latest\n\n# Convert a specific session by ID\ndipeocc convert <session_id> --output-dir projects/claude_code/\n\n# Watch for new sessions and auto-convert them\ndipeocc watch --interval 30 --auto-execute\n\n# View session statistics\ndipeocc stats <session_id>\n```\n\n**Options:**\n- `--format`: Output format (light/native/readable, default: light)\n- `--merge-reads`: Combine consecutive file reads into single nodes\n- `--simplify`: Remove intermediate results for cleaner diagrams\n\nThis feature enables you to automatically transform your Claude Code interactions into reusable DiPeO diagrams, making your AI workflows reproducible and shareable.\n\n#### Export Diagrams to Standalone Python Scripts\n```bash\n# Export a diagram to a Python script that runs without DiPeO\ndipeo export examples/simple_diagrams/simple_iter.light.yaml output.py --light\n\n# Run the exported script directly\npython output.py\n```\n\n**Supported Features:**\n- LLM calls (person_job nodes) → Direct OpenAI/Anthropic API calls\n- Database operations → File I/O\n- Code execution → Inline Python\n- Control flow → Native Python structures\n\nThe exported scripts are self-contained and only require Python 3.10+ and the `openai`/`anthropic` packages. Perfect for deploying DiPeO workflows as standalone applications.\n\nFor detailed usage and examples, see [Diagram-to-Python Export Guide](docs/features/diagram-to-python-export.md).\n\n## Requirements\n- **Python 3.13+** (required for uv support)\n- **Node.js 22+** with **pnpm 10+** (not npm/yarn)\n- **uv** package manager (auto-installed via `make install`)\n- Default LLM: `gpt-5-nano-2025-08-07`\n",
  "bytes": 8921,
  "sha": "f2023e948cad5b5a031fb5267cb951be1be0ba65917a84bc5f40bbba85a316fb",
  "repo_slug": "sorryhyun/dipeo",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/skl_sorryhyun_dipeo_claude_skills_maintain_d_6d857140/readme"
}