Welcome to the Deep Agent Blog

Hey there

I'm starting this blog to document what I'm learning while building Deep Agent. I'll be writing about:

• How the architecture actually works (and why certain decisions were made)
• Practical patterns I've found useful
• New features as they get added
• Things that work well (and things that don't)

What is Deep Agent?

Deep Agent is a way to build more capable AI agents by combining four key capabilities:

1. Planning tools - Agents can break down tasks and track progress with write_todos
2. Virtual filesystem - Persistent file operations that work across multiple tool calls
3. Subagent spawning - Delegate specialized work to focused sub-agents
4. Detailed prompting - Clear system prompts that help agents understand what they can do

This is based on LangChain's DeepAgents framework, reimplemented for Vercel AI SDK v6.

Get Started

If you want to try it out, the Getting Started guide walks through the basics, or you can check out the examples to see it in action.