deep-agents

Deep Agents introduces dynamic subagents that use programmatic orchestration via code scripts instead of tool calls, enabling reliable scaling and complex workflows. The feature integrates with a QuickJS code interpreter for lightweight execution.

The LangChain team has open-sourced the tutorial 'Deep Agents from Scratch', which deconstructs the core design patterns of mainstream agents from scratch, covering task planning, context offloading to a file system, and sub-agent isolation. It includes 5 progressive notebooks, allowing you to build a complete deep research agent hands-on.

Alex, a new LangChain team member, published an article explaining how Deep Agents uses prompt caching to reduce API costs.

An article explaining how to build a Claude Code-like coding agent using LangChain's Deep Agents library, covering the architecture and implementation.

Box Agent, built on LangChain's Deep Agents framework, enables cross-library search, multi-document synthesis, and structured report generation while preserving Box's existing security and permissions model.

A short explanation of Deep Agents by Sydney Runkle, presented by LangChain.

LangChain showcases a macroeconomic research agent built with Deep Agents, LangSmith, and the You.com Finance Research API that analyzes GDP data, detects anomalies, and investigates structural and cyclical drivers at the sector level, producing structured, cited briefings.

LangChain announces Managed Deep Agents in private beta, a hosted API-first runtime for building, running, and operating deep agents in production, leveraging the open-source Deep Agents harness and integrating with LangSmith for durable execution, streaming, and human-in-the-loop workflows.

Deep Agents introduces interpreters: small embedded runtimes that allow agents to write and execute code inside the agent loop, enabling multi-step logic and intermediate state management without full sandbox overhead.

Deep Agents v0.6 introduces a code interpreter, harness profiles for per-model tuning, streaming support, DeltaChannel for checkpoint storage, and ContextHubBackend for versioned agent memory, enabling model-agnostic programmatic tool calling and recursive workflows.

Box and LangChain are showcasing an end-to-end vendor risk analysis solution using Deep Agents and VFS sandboxes, where PDFs are processed by parallel subagents to generate structured reports.

Jacob Lee created an open-source AI coding agent using Deep Agents and ACP that supplants Claude Code, offering multi-model support, LangSmith observability, and human-in-the-loop safeguards.