LangChain ships Deep Agents
LangChain released 'Deep Agents', a structured runtime that formalizes planning, persistent memory, and context isolation for multi‑step agents — traits aimed at making complex agent workflows easier to debug and scale reported.
DeepAgents is published as deepagents 0.4.10 on PyPI, with the package listing a Mar 11, 2026 release and PyPI recommending Python 3.11+ for installation. pypi.org The runtime compiles graphs into a CompiledStateGraph that supports streaming invocations and optional superstep checkpointing for state restore; the LangGraph API explicitly documents checkpoint hooks on CompiledStateGraph. reference.langchain.com Native tracing and evaluation hook-ups are provided via LangSmith, and the Deep Agents docs show how to enable LangSmith tracing (requiring a LangSmith API key) so every node, LLM call, and tool result can be inspected in the LangSmith UI. docs.langchain.com Reliability middleware in the shipped SDK includes auto-summarization of histories, large tool-result eviction, and pluggable backends (local/ephemeral/persistent/sandbox) alongside built-in filesystem and shell tools (read_file, write_file, edit_file, ls, grep). langchain.com Developer experience components include an SDK plus a deepagents-cli with an --acp mode to run an ACP server for editor integration, active CLI releases on GitHub, and the project hosted as open-source on the langchain-ai GitHub repo. github.com Early enterprise signals: LangChain reported its internal GTM system built on Deep Agents produced a 250% increase in lead-to-qualified-opportunity conversion and freed 1,320 hours monthly for sales between Dec 2025–Mar 2026, while partners like Box have published integration guides showing document-backed multi-agent workflows. blockchain.news
