Skip to content

About

An AI-powered agent builder that transforms natural language instructions into executable visual workflows. It generates, validates, tests, and automatically repairs AI agent workflows using LLMs and LangGraph.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

8 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

AgentForge AI 🛠️⚡

AgentForge AI is an AI-powered visual agent builder that converts natural language instructions into executable visual workflows. It automatically generates workflows, validates topology, tests functionality, detects failures, and attempts autonomous self-repairs.


🌟 Technology Stack

  • Frontend: React 19, TypeScript, Vite, Tailwind CSS, React Flow (@xyflow/react), Lucide React
  • Backend: Python 3.11, FastAPI, Pydantic v2, SQLite, SQLAlchemy 2.0
  • AI & Reasoning Layer: Google Gemini API, LangGraph (StateGraph runtime)
  • Testing: Pytest, Pytest-Asyncio, HTTPX

📁 Monorepo Architecture

agent_workflow/
├── frontend/             # React + Vite visual workflow editor
├── backend/              # FastAPI + SQLAlchemy SQLite API service
│   ├── app/
│   │   ├── api/v1/       # Endpoints (including /api/v1/health)
│   │   ├── core/         # App settings & environment configs
│   │   ├── database/     # SQLAlchemy engine & session factory
│   │   ├── models/       # Database models (Workflow, Execution)
│   │   ├── schemas/      # Pydantic schemas
│   │   └── services/     # Business logic & workflow services
│   └── tests/            # Pytest test suite
├── shared/               # Single Source of Truth
│   ├── workflow.schema.json  # Workflow JSON Schema (Draft 2020-12)
│   ├── node_definitions.json # 7 canonical node type catalog
│   └── types.ts              # Shared TypeScript definitions
├── docs/                 # Engineering Documentation
│   ├── ARCHITECTURE.md       # High-level diagrams & self-repair loop
│   ├── API_SPECIFICATION.md  # REST API specification & schemas
│   ├── DEVELOPMENT_GUIDE.md  # 6-member team guide & conventions
│   └── README.md
├── .env.example          # Environment variables template (No secrets)
├── .gitignore            # Git exclusion rules
└── README.md             # This document

👥 Six-Member Engineering Team Responsibilities

# Role / Member Key Responsibilities Primary Ownership
1 Frontend Development Visual drag-and-drop canvas with React Flow, node parameter sidebars, execution status indicators, and responsive UI. frontend/
2 Backend and Database RESTful routing, SQLite/SQLAlchemy ORM models, migrations, CRUD endpoints, and database connection lifecycle. backend/app/api/, backend/app/database/, backend/app/models/
3 AI Agent Architect Google Gemini API integration, prompt orchestration, translating natural language prompts into valid structured workflow graphs. backend/app/services/ai/
4 Workflow Generation & Components Graph topology validation, cycle detection, input/output port compatibility, component catalog, schema versioning. shared/, backend/app/services/workflow/
5 Workflow Execution Engine Compiling workflow graphs to LangGraph StateGraph, asynchronous node runner, tool execution sandbox, human approval gates. backend/app/services/engine/
6 Testing, Validation & Self-Repair End-to-end Pytest test suite, failure diagnosis agent, automated graph patching and self-repair verification loop. backend/tests/, backend/app/services/repair/

🧩 Supported Node Types

The 7 standard node primitives supported by AgentForge AI (defined in shared/node_definitions.json):

  1. Input (FileInput): Ingests user prompts, payloads, or webhook triggers.
  2. LLM (Bot): Executes prompts and reasoning chains with Google Gemini models.
  3. Document Reader (FileText): Extracts and chunks text from PDF, DOCX, Markdown, and text context.
  4. Tool (Wrench): Performs API calls, search queries, or custom Python actions.
  5. Condition (GitFork): Branching logic evaluating conditional expressions.
  6. Human Approval (UserCheck): Human-in-the-loop pause gate requiring user sign-off.
  7. Output (FileOutput): Terminal node returning formatted results, reports, and metrics.

🚀 Quickstart Guide

1. Prerequisites

  • Node.js: v20.x or v22.x
  • npm: v10.x or higher
  • Python: v3.11.x

2. Configure Environment

# On Windows PowerShell:
Copy-Item .env.example .env

# On macOS / Linux:
cp .env.example .env

(Do not commit your .env or real API key).

3. Start Backend

cd backend

# Create virtual environment (if not already created)
python -m venv .venv

# Activate virtual environment
# Windows PowerShell:
.venv\Scripts\Activate.ps1
# macOS / Linux:
source .venv/bin/activate

# Install dependencies
pip install -r requirements.txt

# Run automated tests
pytest -v

# Start FastAPI server
uvicorn app.main:app --reload --port 8000
  • API Base: http://localhost:8000/api/v1
  • Health Check: http://localhost:8000/api/v1/health
  • Interactive API Docs: http://localhost:8000/docs

4. Start Frontend

In a separate terminal:

cd frontend

# Install dependencies
npm install

# Start Vite dev server
npm run dev
  • Frontend UI: http://localhost:5173

📚 Documentation Links

About

An AI-powered agent builder that transforms natural language instructions into executable visual workflows. It generates, validates, tests, and automatically repairs AI agent workflows using LLMs and LangGraph.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages