AgentStack

AgentStack is an open-source, MIT-licensed command-line tool from the agentstack-ai organization that generates the starting structure for an AI agent project. It scaffolds the framework, LLM, tooling, and observability layers so developers can begin writing agent code. The latest listed release is 0.3.7, published 2025-07-10.

What It Is

AgentStack is a project scaffolding and code-generation CLI for agent development, similar in spirit to create-react-app. Running agentstack init <project_name> creates a project directory with an initial structure and installs dependencies. The README states it is not a low-code alternative, so developers still need to understand their chosen agent framework.

Workflow

After initialization, developers generate agents and tasks with agentstack generate agent/task <name> and edit the agents.yaml and tasks.yaml files in src/config. Tools are added with agentstack tools add, and agentstack run runs the project in development mode. Pre-built templates are available as a shortcut. Production deployment support is described as coming soon.

Supported Stack

The README lists LLM support through LiteLLM or LangChain, and frameworks including CrewAI, LangGraph, OpenAI Swarms, and LlamaStack. Pydantic AI, Eliza, AG2, and Autogen are on the roadmap. AgentOps observability is included by default, and the project maintains a framework-agnostic tool repository.

Design Principles

The project emphasizes no required configuration, no lock-in, and provider agnosticism across stack layers. Roadmap items include a prompting layer (DSPy), an eval layer, a UI layer, generated testing, and integrated benchmarking.