AgentStack

Create AI agent projects from the command line.

• Quickstart Guide – How to create a new agent project.
• Video Tutorial – Follow along and build a web scrape agent with AgentStack

AgentStack works on macOS, Windows, and Linux.
If something doesn't work, please file an issue.
If you have questions or need help, please ask in our Discord community.

🛠️🚨 AgentStack is in open preview. We're building in public, use at your own risk but have fun :)

AgentStack serves as a great tool for starting your agent project and offers many CLI utilities for easy code-gen throughout the development process.

AgentStack is not a low-code alternative to development. Developers will still need an understanding of how to build with their selected agent framework.

Quick Overview

pip install agentstack
agentstack init <project_name>

Get Started Immediately

You don't need to install or configure tools like LangChain or LlamaIndex.
They are preconfigured and hidden so that you can focus on the code.

Create a project, and you're good to go.

Creating an Agent Project

You'll need to have Python 3.10+ on your local development machine (but it's not required on the server). We recommend using the latest version. You can use pyenv to switch Python versions between different projects.

To create a new agent project, run:

pip install agentstack
agentstack init

It will create a directory with your project name inside the current folder.
Inside that directory, it will generate the initial project structure and install the transitive dependencies.

No configuration or complicated folder structures, only the files you need to build your agent project.
Once the initialization is done, you can open your project folder:

cd my-agent-project

Building Agent Functionality

After generating a project, the next step is to build your agent project by creating Agents and Tasks. You can do this quickly with AgentStack:

agentstack generate agent/task <name>

Modify the agents and tasks by changing the agents.yaml and tasks.yaml configuration files in src/config

Tooling

One of AgentStack's core principles is to establish the de facto agent stack. A critical component of this stack is the tooling and functionality given to agents beyond simply LLM capabilities.

AgentStack has worked to make access to tools as easy as possible, staying framework agnostic and featuring the best tools.

A list of all tools can be found on our docs.

Adding tools is as simple as

agentstack tools add <tool_name>

Running Your Agent

python src/main.py

Runs the agent project in development mode.

👀 Support for easy production deployment of agents is coming soon.

Philosophy

• Agents should be easy: There are so many frameworks out there, but starting from scratch is a pain. Similar to create-react-app, AgentStack aims to simplify the "from scratch" process by giving you a simple boilerplate of an agent. It uses popular agent frameworks and LLM providers, but provides a cohesive curated experience on top of them.

• No Configuration Required: You don't need to configure anything. A reasonably good configuration of both development and production builds is handled for you so you can focus on writing code.

• No Lock-In: You can customize your setup at any time. AgentStack is designed to make it easy to get the components you need running right off the bat; it's up to you what to do next.

AgentStack is not designed to be a low-code solution to building agents. Instead it is a great head-start for starting an agent project from scratch.

Roadmap

Frameworks

CrewAI

Development of AgentStack is being done primarily on CrewAI.

AutoGen

Some work has been done to add Microsoft's AutoGen, although these efforts have been paused. AutoGen is currently in the process of making large design decisions that will effect the integration with AgentStack.

Tools

• Core Tools built by AgentStack
• Preferred partners in the package directly
• Community partner tools added through external repositories

Other Features

• Generated testing
• Integrated benchmarking
• Easy integration of tools for browsing, RAG, and more.
• A fast interactive test runner with built-in support for coverage reporting.
• A live development server that warns about common mistakes.
• A build script to bundle your project for production.
• Integration with AgentOps for AI agent observability.
• Hassle-free updates for the above tools with a single dependency.

License

AgentStack is open source software licensed as MIT.

How to Contribute

AgentStack is a new project built by passionate AI agent developers! We'd love help making this tool better. Easy first issues are available, create new issues with feature ideas, or chat with us on our Discord.

If you are an Agent Tool developer, feel free to create an issue or even a PR to add your tool to AgentStack.