Changelog

Newest on top.

2026-07-01

Monthly check of LangChain, LangGraph, Deep Agents, and LangSmith. Source: the June 2026 newsletter and linked docs/blogs.

Deep Agents — Rubrics. New RubricMiddleware (beta) adds a self-check loop: you pass a rubric at invoke time, a grader sub-agent scores the run per criterion, and the agent keeps fixing until it passes or hits max_iterations. Updated Middleware and Deep Agents. Source: blog, docs.
Deep Agents — programmatic subagents. You can now spin up subagents from inside interpreter code to fan out work and merge results, not just via the task tool. Updated Multi-agent. Source: docs.
LangSmith — new Fleet + tracing features. Fleet On-Call Copilot template (alert triage), Computer Use in Fleet (isolated virtual machine), a new voice-trace UI with an inline audio player, live experiment status tracking, and native Slack notifications for Engine Issues. Updated Observability. Source: June newsletter.
New course. "Introduction to LangSmith Deployment" on LangChain Academy. Added to reading list.

Added a new topic: Structured output & validation. Covers ToolStrategy vs ProviderStrategy, Pydantic schemas, handle_errors retries, and validating tool calls, tool outputs, and model outputs.
Expanded the full guide with a "Structured output & validation" section plus three new appendix snippets.
Added a new topic: Multimodal agents — text + image (also audio/video). Covers content blocks, image URL/base64/file_id input, and reading image output.

Converted the whole knowledge base to HTML: a main hub plus a standalone page per topic.
Created folder structure and starter pages for every topic.
Added the full one-page guide: langchain-agents-guide.html (v1.0).
Based on LangChain 1.x / LangGraph 1.x.