6 posts
Deep Agents SDK is an agent framework from LangChain that addresses the missing planning, memory, and context management found in typical agent frameworks. This article explains its relationship with LangGraph and LangChain, and when you actually need it.
Deep Agents' write_todos tool lets the agent decompose a task into steps and track progress before execution begins. This article explains the mechanics of write_todos and how it transforms complex task execution from 'doing things haphazardly' into 'advancing with a plan.'
Deep Agents includes a virtual filesystem, letting the agent use ls, read_file, write_file, and edit_file to manage task data without stuffing all intermediate results into the context window. This article explains the design logic and usage of the filesystem tools.
Deep Agents' task tool lets the main agent spawn subagents and hand off specific subtasks — each subagent has independent context so it doesn't contaminate the main agent. This article explains the design logic of subagents and how to use them in practice.
Deep Agents uses LangGraph Memory Store to implement cross-conversation long-term memory, letting the agent recall user preferences, project state, and completed work in the next session. This article covers setup and the design strategy for memory management.
Deep Agents CLI is a concrete implementation of the SDK — a coding agent that runs in your terminal, with support for Skills, long-term memory, and running code in a sandbox. This article covers installation, configuration, and real-world usage scenarios.