deerflow2/CLAUDE.md

# CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

## Project Overview

DeerFlow is an open-source **AI super agent harness** built on LangGraph/LangChain. It's a full-stack application that orchestrates sub-agents, memory, sandboxes, and extensible skills to perform complex multi-step tasks.

**Architecture**:
- **LangGraph Server** (port 2024): Agent runtime and workflow execution
- **Gateway API** (port 8001): FastAPI REST API for models, MCP, skills, memory, artifacts, uploads
- **Frontend** (port 3000): Next.js 16 web interface
- **Nginx** (port 2026): Unified reverse proxy entry point
- **Provisioner** (port 8002, optional): Started only when sandbox is configured for provisioner/Kubernetes mode

**Key Technologies**:
- **Backend**: Python 3.12+, LangGraph/LangChain, FastAPI, uv package manager
- **Frontend**: Next.js 16, React 19, TypeScript 5.8, Tailwind CSS 4, pnpm
- **AI/ML**: Model-agnostic (any OpenAI-compatible API), multi-model support with thinking/vision capabilities

## Commands

### Root Directory (Full Application)
```bash
make check      # Check system requirements (Node.js 22+, pnpm, uv, nginx)
make config     # Generate local configuration files from templates
make install    # Install all dependencies (frontend + backend)
make setup-sandbox # Pre-pull sandbox container image (recommended for Docker sandbox)
make dev        # Start all services (LangGraph + Gateway + Frontend + Nginx) on localhost:2026
make stop       # Stop all running services
make clean      # Clean up processes and temporary files

# Docker Development (Recommended)
make docker-init    # Build custom k3s image with pre-cached sandbox image
make docker-start   # Start Docker services (mode-aware from config.yaml)
make docker-stop    # Stop Docker development services
make docker-logs    # View Docker development logs
```

### Backend Directory (`/backend/`)
```bash
make install    # Install backend dependencies with uv
make dev        # Run LangGraph server only (port 2024)
make gateway    # Run Gateway API only (port 8001)
make lint       # Lint with ruff
make format     # Format code with ruff
uv run pytest   # Run backend tests
```

### Frontend Directory (`/frontend/`)
```bash
pnpm dev        # Dev server with Turbopack (http://localhost:3000)
pnpm build      # Production build
pnpm check      # Lint + type check (run before committing)
pnpm lint       # ESLint only
pnpm lint:fix   # ESLint with auto-fix
pnpm typecheck  # TypeScript type check (tsc --noEmit)
pnpm start      # Start production server
```

## Architecture

### Microservices Architecture
```
┌──────────────────────────────────────┐
│          Nginx (Port 2026)           │
│      Unified reverse proxy           │
└───────┬──────────────────┬───────────┘
        │                  │
/api/langgraph/*  │                  │  /api/* (other)
        ▼                  ▼
 ┌────────────────────┐  ┌────────────────────────┐
 │ LangGraph Server   │  │   Gateway API (8001)   │
 │    (Port 2024)     │  │   FastAPI REST         │
 │                    │  │                        │
 │ ┌────────────────┐ │  │ Models, MCP, Skills,   │
 │ │  Lead Agent    │ │  │ Memory, Uploads,       │
 │ │  ┌──────────┐  │ │  │ Artifacts              │
 │ │  │Middleware│  │ │  └────────────────────────┘
 │ │  │  Chain   │  │ │
 │ │  └──────────┘  │ │
 │ │  ┌──────────┐  │ │
 │ │  │  Tools   │  │ │
 │ │  └──────────┘  │ │
 │ │  ┌──────────┐  │ │
 │ │  │Subagents │  │ │
 │ │  └──────────┘  │ │
 │ └────────────────┘ │
 └────────────────────┘
```

### Core Components

**Lead Agent** (`src/agents/lead_agent/agent.py`):
- Entry point: `make_lead_agent(config: RunnableConfig)` registered in `langgraph.json`
- Dynamic model selection via `create_chat_model()` with thinking/vision support
- Tools loaded via `get_available_tools()` - combines sandbox, built-in, MCP, community, and subagent tools

**Middleware Chain** (11 middlewares in strict order):
1. **ThreadDataMiddleware** - Creates per-thread isolated directories
2. **UploadsMiddleware** - Tracks and injects newly uploaded files
3. **SandboxMiddleware** - Acquires sandbox, stores `sandbox_id` in state
4. **DanglingToolCallMiddleware** - Injects placeholder ToolMessages for interrupted tool calls
5. **SummarizationMiddleware** - Context reduction when approaching token limits
6. **TodoListMiddleware** - Task tracking with `write_todos` tool (plan mode)
7. **TitleMiddleware** - Auto-generates thread title
8. **MemoryMiddleware** - Queues conversations for async memory update
9. **ViewImageMiddleware** - Injects base64 image data before LLM call (vision support)
10. **SubagentLimitMiddleware** - Truncates excess `task` tool calls to enforce concurrency limits
11. **ClarificationMiddleware** - Intercepts `ask_clarification` tool calls, interrupts via `Command(goto=END)`

**Sandbox System** (`src/sandbox/`):
- **Interface**: Abstract `Sandbox` with `execute_command`, `read_file`, `write_file`, `list_dir`
- **Provider Pattern**: `SandboxProvider` with `acquire`, `get`, `release` lifecycle
- **Implementations**: `LocalSandboxProvider` (local filesystem), `AioSandboxProvider` (Docker-based isolation)
- **Virtual Path System**: Agent sees `/mnt/user-data/{workspace,uploads,outputs}`, `/mnt/skills` mapped to physical paths

**Subagent System** (`src/subagents/`):
- **Built-in Agents**: `general-purpose` (all tools except `task`) and `bash` (command specialist)
- **Concurrency**: `MAX_CONCURRENT_SUBAGENTS = 3` enforced by `SubagentLimitMiddleware`
- **Execution**: Dual thread pool - `_scheduler_pool` (3 workers) + `_execution_pool` (3 workers)
- **Flow**: `task()` tool → `SubagentExecutor` → background thread → poll 5s → SSE events → result

**Memory System** (`src/agents/memory/`):
- **Components**: `updater.py` (LLM-based memory updates), `queue.py` (debounced update queue), `prompt.py` (templates)
- **Data Structure**: Stored in `backend/.deer-flow/memory.json` with user context, history, and facts
- **Workflow**: MemoryMiddleware filters messages → queue debounces (30s) → background LLM extracts updates → atomic file I/O

**Tool System** (`src/tools/`):
- **Config-defined tools**: Resolved from `config.yaml` via `resolve_variable()`
- **MCP tools**: From enabled MCP servers (lazy initialized, cached with mtime invalidation)
- **Built-in tools**: `present_files`, `ask_clarification`, `view_image` (vision support)
- **Subagent tool**: `task` (delegate to subagent)
- **Community tools**: `tavily/` (web search/fetch), `jina_ai/` (web fetch), `firecrawl/` (scraping), `image_search/` (DuckDuckGo)

**Skills System** (`src/skills/`):
- **Location**: `deer-flow/skills/{public,custom}/`
- **Format**: Directory with `SKILL.md` (YAML frontmatter: name, description, license, allowed-tools)
- **Loading**: `load_skills()` scans directories, parses SKILL.md, reads enabled state from extensions_config.json
- **Installation**: `POST /api/skills/install` extracts .skill ZIP archive to custom/ directory

### Configuration System

**Main Configuration** (`config.yaml` in project root):
- Setup: Copy `config.example.yaml` to `config.yaml`
- Configuration priority: explicit `config_path` → `DEER_FLOW_CONFIG_PATH` env → `config.yaml` in current dir → `config.yaml` in parent dir
- Config values starting with `$` are resolved as environment variables (e.g., `$OPENAI_API_KEY`)

**Extensions Configuration** (`extensions_config.json` in project root):
- Contains MCP servers and skills configuration
- Configuration priority: explicit `config_path` → `DEER_FLOW_EXTENSIONS_CONFIG_PATH` env → `extensions_config.json` in current dir → `extensions_config.json` in parent dir

**Key Config Sections**:
- `models[]` - LLM configs with `use` class path, `supports_thinking`, `supports_vision`
- `tools[]` - Tool configs with `use` variable path and `group`
- `tool_groups[]` - Logical groupings for tools
- `sandbox.use` - Sandbox provider class path
- `skills.path` / `skills.container_path` - Host and container paths to skills directory
- `title` - Auto-title generation
- `summarization` - Context summarization
- `subagents.enabled` - Master switch for subagent delegation
- `memory` - Memory system configuration

### Gateway API (`src/gateway/`)

FastAPI application on port 8001 with health check at `GET /health`.

**Routers**:
- **Models** (`/api/models`): `GET /` (list models), `GET /{name}` (model details)
- **MCP** (`/api/mcp`): `GET /config` (get config), `PUT /config` (update config)
- **Skills** (`/api/skills`): `GET /` (list), `GET /{name}` (details), `PUT /{name}` (update enabled), `POST /install` (install from .skill archive)
- **Memory** (`/api/memory`): `GET /` (memory data), `POST /reload` (force reload), `GET /config` (config), `GET /status` (config + data)
- **Uploads** (`/api/threads/{id}/uploads`): `POST /` (upload files with auto-conversion), `GET /list` (list), `DELETE /{filename}` (delete)
- **Artifacts** (`/api/threads/{id}/artifacts`): `GET /{path}` (serve artifacts), `?download=true` for file download

### Frontend Architecture (`/frontend/`)

**Source Layout** (`src/`):
- **`app/`** - Next.js App Router: `/` (landing), `/workspace/chats/[thread_id]` (chat)
- **`components/`** - React components: `ui/` (Shadcn UI), `ai-elements/` (Vercel AI SDK), `workspace/` (chat), `landing/` (landing)
- **`core/`** - Business logic: `threads/` (state management), `api/` (LangGraph client), `artifacts/` (loading/caching), `i18n/`, `settings/`, `memory/`, `skills/`, `messages/`, `mcp/`, `models/`
- **`hooks/`** - Shared React hooks
- **`lib/`** - Utilities (`cn()` from clsx + tailwind-merge)
- **`server/`** - Server-side code (better-auth, not yet active)
- **`styles/`** - Global CSS with Tailwind v4 `@import` syntax

**Data Flow**:
1. User input → thread hooks (`core/threads/hooks.ts`) → LangGraph SDK streaming
2. Stream events update thread state (messages, artifacts, todos)
3. TanStack Query manages server state; localStorage stores user settings
4. Components subscribe to thread state and render updates

## Development Workflow

### Running the Full Application
From the **project root** directory:
```bash
make dev
```
This starts all services and makes the application available at `http://localhost:2026`.

**Nginx routing**:
- `/api/langgraph/*` → LangGraph Server (2024)
- `/api/*` (other) → Gateway API (8001)
- `/` (non-API) → Frontend (3000)

### Running Services Separately
From the **backend** directory:
```bash
# Terminal 1: LangGraph server
make dev

# Terminal 2: Gateway API
make gateway
```

Direct access (without nginx):
- LangGraph: `http://localhost:2024`
- Gateway: `http://localhost:8001`

### Docker Development (Recommended)
```bash
make docker-init    # Build images and install dependencies (first time)
make docker-start   # Start all services with hot-reload
```
Access at `http://localhost:2026`. Services automatically restart on code changes.

## Key Features

### File Upload
- Endpoint: `POST /api/threads/{thread_id}/uploads`
- Supports: PDF, PPT, Excel, Word documents (auto-converted via `markitdown`)
- Files stored in thread-isolated directories
- Agent receives uploaded file list via `UploadsMiddleware`

### Plan Mode
- Controlled via runtime config: `config.configurable.is_plan_mode = True`
- Provides `write_todos` tool for task tracking
- One task in_progress at a time, real-time updates

### Context Summarization
- Automatic conversation summarization when approaching token limits
- Configured in `config.yaml` under `summarization` key
- Trigger types: tokens, messages, or fraction of max input
- Keeps recent messages while summarizing older ones

### Vision Support
- For models with `supports_vision: true`
- `ViewImageMiddleware` processes images in conversation
- `view_image_tool` added to agent's toolset
- Images automatically converted to base64 and injected into state

## Code Style

### Backend (Python)
- Uses `ruff` for linting and formatting
- Line length: 240 characters
- Python 3.12+ with type hints
- Double quotes, space indentation

### Frontend (TypeScript)
- **Imports**: Enforced ordering (builtin → external → internal → parent → sibling), alphabetized
- **Unused variables**: Prefix with `_`
- **Class names**: Use `cn()` from `@/lib/utils` for conditional Tailwind classes
- **Path alias**: `@/*` maps to `src/*`
- **Components**: `ui/` and `ai-elements/` are generated from registries - don't manually edit these

## Testing

### Backend Tests
```bash
cd backend
uv run pytest
```
**Regression tests** (run in CI for every PR):
- `tests/test_docker_sandbox_mode_detection.py` - mode detection from `config.yaml`
- `tests/test_provisioner_kubeconfig.py` - kubeconfig file/directory handling

### Frontend Tests
No test framework is currently configured.

## Documentation Update Policy

**CRITICAL: Always update README.md and CLAUDE.md after every code change**

When making code changes, you MUST update the relevant documentation:
- Update `README.md` for user-facing changes (features, setup, usage instructions)
- Update `CLAUDE.md` for development changes (architecture, commands, workflows, internal systems)
- Keep documentation synchronized with the codebase at all times
- Ensure accuracy and timeliness of all documentation

## Project Structure

```
deer-flow/
├── Makefile                    # Root commands (check, install, dev, stop)
├── config.yaml                 # Main application configuration
├── extensions_config.json      # MCP servers and skills configuration
├── backend/                    # Backend application
│   ├── Makefile               # Backend-only commands (dev, gateway, lint)
│   ├── langgraph.json         # LangGraph server configuration
│   ├── src/
│   │   ├── agents/            # LangGraph agent system
│   │   │   ├── lead_agent/    # Main agent (factory + system prompt)
│   │   │   ├── middlewares/   # 11 middleware components
│   │   │   ├── memory/        # Memory extraction, queue, prompts
│   │   │   └── thread_state.py # ThreadState schema
│   │   ├── gateway/           # FastAPI Gateway API
│   │   │   ├── app.py         # FastAPI application
│   │   │   └── routers/       # 6 route modules
│   │   ├── sandbox/           # Sandbox execution system
│   │   │   ├── local/         # Local filesystem provider
│   │   │   ├── sandbox.py     # Abstract Sandbox interface
│   │   │   ├── tools.py       # bash, ls, read/write/str_replace
│   │   │   └── middleware.py  # Sandbox lifecycle management
│   │   ├── subagents/         # Subagent delegation system
│   │   │   ├── builtins/      # general-purpose, bash agents
│   │   │   ├── executor.py    # Background execution engine
│   │   │   └── registry.py    # Agent registry
│   │   ├── tools/builtins/    # Built-in tools (present_files, ask_clarification, view_image)
│   │   ├── mcp/               # MCP integration (tools, cache, client)
│   │   ├── models/            # Model factory with thinking/vision support
│   │   ├── skills/            # Skills discovery, loading, parsing
│   │   ├── config/            # Configuration system (app, model, sandbox, tool, etc.)
│   │   ├── community/         # Community tools (tavily, jina_ai, firecrawl, image_search, aio_sandbox)
│   │   ├── reflection/        # Dynamic module loading (resolve_variable, resolve_class)
│   │   └── utils/             # Utilities (network, readability)
│   ├── tests/                 # Test suite
│   └── docs/                  # Documentation
├── frontend/                  # Next.js frontend application
│   ├── src/
│   │   ├── app/              # Next.js App Router
│   │   ├── components/       # React components
│   │   ├── core/             # Business logic
│   │   ├── hooks/            # Shared React hooks
│   │   ├── lib/              # Utilities
│   │   ├── server/           # Server-side code
│   │   └── styles/           # Global CSS with Tailwind v4
│   └── package.json          # Next.js 16, React 19, TypeScript 5.8, pnpm
└── skills/                    # Agent skills directory
    ├── public/               # Public skills (committed)
    └── custom/               # Custom skills (gitignored)
```