Middleware

Deep agents use a modular middleware architecture where each core capability is implemented as composable middleware. This design allows you to customize agent behavior by adding, removing, or modifying middleware.

Learn more: For architecture details and customization patterns, see the Middleware documentation.

Built-in Middleware

TodoListMiddleware

Provides the write_todos tool for task planning and progress tracking. Essential for complex, multi-step tasks.

FilesystemMiddleware

Provides tools for context management through file operations. This is one of the main challenges in building effective agents—the filesystem middleware helps agents work with large amounts of context without overflowing the context window.

Tool	Description
ls	List files in a directory
read_file	Read file contents
write_file	Write content to a file
edit_file	Edit an existing file
glob	Find files matching a pattern
grep	Search for text within files
execute	Run shell commands (sandbox backends only)

SubAgentMiddleware

Provides the task tool for spawning specialized subagents. Subagents are useful for:

• Context isolation: Keep the main agent's context clean
• Specialization: Give subagents specific tools and prompts for focused tasks
• Parallel work: Delegate independent subtasks

MemoryMiddleware

Enables long-term memory capabilities using LangGraph Store.

Learn more: See Long-term Memory for usage patterns.

SummarizationMiddleware

Automatically summarizes long conversations to prevent context overflow.

SkillsMiddleware

Loads reusable skill modules to extend agent capabilities.

Learn more: See Skills for creating and loading skills.

API Reference