Workflow Platform

English | 中文

A personal AI workflow engine built on Eino. Design workflows visually, run them with AI-powered agents that can search the web, call APIs, execute scripts, and reason through multi-step tasks.

Features

• Visual workflow builder — drag-and-drop nodes, connect with Flow/Data edges, resizable properties panel
• AI Agent — eino ReAct agent with multi-step reasoning, tool calling, and permission controls
• Built-in tools — now, web_search, web_fetch, read_file, write_file — toggle permissions per agent
• Node-based tools — API Call and Code nodes can be called as tools by the Agent
• Extractor — upload files, LLM summarizes, passes context to Agent via Data edges
• Code execution — Code nodes execute JS/Python scripts with data variable
• Cron scheduling — Schedule-type templates run automatically on cron expressions
• Dual edge system — Flow edges for execution order, Data edges for context sharing
• Undo/Redo — Full undo/redo support with keyboard shortcuts (Ctrl+Z / Ctrl+Y)
• Template management — Card grid UI with hover menu, create/edit/delete templates
• Instance tracking — Real-time node state tracking (pending/running/success/failed) with expandable logs
• Markdown rendering — Agent responses rendered as formatted Markdown (headings, tables, code blocks)
• LLM profile selection — Each Agent and Extractor node selects its own LLM profile (provider, model, API key)
• Sub-workflow — Call other saved templates as sub-processes

Node Types

Type	Label	Color	Description
call	API Call	Blue	HTTP request to external API
agent	AI Agent	Purple	LLM-powered ReAct agent with tool calling
condition	Condition	Orange	IF/ELSE branching with expression evaluation
code	Code	Green	JS/Python script execution
extractor	Extractor	Cyan	File upload + LLM summarization
filter	Filter	Teal	Data filtering/transformation
subworkflow	Sub-Workflow	Indigo	Call another template as a sub-process

Edge Types

Type	Description
flow	Execution order — output port → input port
data	Context sharing — data port (bottom) → target node

Condition Node

Condition nodes have two labeled output ports:

• IF (green) — route taken when expression is true
• ELSE (red) — route taken when expression is false

Expressions use state. prefix, e.g. state._global.score >= 60. Supported operators: >=, <=, >, <, =, !=.

Agent Permissions

Each Agent node has independent permission toggles:

• Read local files — enables read_file tool (read/list/search files)
• Write local files — enables write_file tool
• Web access — enables web_search + web_fetch tools

Quick Start

Prerequisites

• Go 1.22+
• Node.js 18+ (for frontend development)

Build & Run

# Build frontend (optional, static files already included)
cd web && npm install && npm run build && cd ..

# Build and run server
go build -ldflags=-checklinkname=0 -o workflow-server ./cmd/server/
./workflow-server

Open http://localhost:8080

Frontend Development

cd web
npm install
npm run dev   # Vite dev server on port 5173, proxies API to :8080

Configuration

Edit config.yaml:

server:
  port: 8080

database:
  path: "workflow.db"    # SQLite

llm:
  profiles:
    - name: my-llm
      provider: openai
      model: gpt-4o
      api_key: sk-xxx
      base_url: "https://api.openai.com"

Architecture

workflow-platform/
├── cmd/server/          # Entry point
├── internal/
│   ├── api/             # HTTP handlers
│   ├── config/          # Config loading
│   ├── engine/          # Workflow execution engine (eino)
│   ├── model/           # Data types
│   ├── server/          # HTTP server + embedded static files
│   └── store/           # SQLite/Memory storage
├── web/                 # React + TypeScript frontend
│   ├── src/
│   │   ├── pages/       # TemplatesPage, InstancesPage, BuilderPage
│   │   ├── components/  # Reusable UI components
│   │   ├── store.ts     # Zustand state management
│   │   └── types.ts     # TypeScript type definitions
│   └── dist/            # Build output → copied to static/
└── config.yaml          # Server + LLM configuration

Tech Stack

• Backend: Go, Eino (workflow engine)
• Frontend: React 18, TypeScript, Vite, Zustand
• Storage: SQLite
• LLM: OpenAI-compatible API (DeepSeek, OpenAI, etc.)