AgentSmith

AgentSmith is a developer-first, visual framework for building, testing, and exporting LangGraph-based AI agents.

Design intelligent workflows using a drag-and-drop interface, describe flows in natural language, define custom state/message schemas, and generate full runnable Python code. Whether you're using OpenAI, LLaMA, or your own tools, AgentSmith helps you go from idea → graph → working agent in seconds.

“Your agent. Your logic. Your code.”

---

✨ Features

• 🧩 Drag & Drop Agent Builder
  Design nodes, edges, and async flows using a canvas powered by React Flow.

• 🧠 Typed State Definition (LangGraph-native)
  Visually define TypedDict-based agent state and message schemas.

• 🚀 Python / LangGraph Code Export
  Export your agent as a standalone Python module using LangGraph. Then just copy paste the code that works out of the box.

• 🧬 Per-Node Code Editing
  Modify each node’s logic directly in a Monaco (VSCode-style) editor.

• 🔌 Modular LLM Backend
  Built-in support for OpenAI, Anthropic, Hugging Face Transformers, and local LLMs, currently Llama.cpp (more to come...).

• 🛠️ Custom Tool Creation
  Define and reuse tools via code or natural language.

• 🤖 Chatbot
  Test your LLMs in a built-in chat interface with debig options and parameter tuning.

• 🧬 LLM Fine-Tuning
  Fine-tune your LLMs on your own datasets using a built-in interface.

---

💡 Why AgentSmith?

AgentSmith is built for developers and advanced users who want:

• ✅ Real state management via LangGraph
• ✅ Full Python code and version control
• ✅ Pluggable local/remote LLMs
• ✅ Automated structured LLM outputs for models that do not support it (Llama.cpp etc.) by integrating outlines
• ✅ Typed agent state, message schema, and input/output mapping

---

🛠️ Tech Stack

Layer	Tech Stack
Frontend	React + React Flow + TailwindCSS + Monaco Editor
Backend	Python + FastAPI + LangChain + LangGraph
LLM APIs	OpenAI, Anthropic, HuggingFace, Llama.cpp
Local LLMs	OpenAI, Anthropic, HuggingFace, Llama.cpp
Storage	SQLite / TinyDB / Local JSON

---

📦 Coming Soon

• 🔍 Flow testing with input/output tracing
• 💾 Flow versioning & Git-based exports
• 🧰 Tool and flow marketplace
• 🧪 Integrated local model benchmarking
• ☁️ Cloud sync & deploy

---

🏗️ Architecture

---

📸 Screenshots

Coming soon: animated walkthroughs of building agents, customizing state, and exporting LangGraph code.

---

🚀 Quickstart

Full instructions will be added soon — includes local dev, docker setup, and example flows.

$ git clone https://github.com/nMaroulis/agent-smith
$ cd agent-smith

# Backend
$ cd backend
$ uv venv .venv
$ source .venv/bin/activate
$ uv pip install -r requirements.txt
$ python main.py

# Frontend
$ cd frontend
$ npm install
$ npm run dev

🤝 Contributing

I welcome contributions, ideas, and extensions. AgentSmith is modular by design — whether you’re adding a new LLM provider or a UI feature, I’d love your input.

📄 License

MIT — free for personal and commercial use.