LangChain AI Updates: October 23, 2026

1. Deep Agents Talon 0.1.0 Adds Effort Controls, an NVIDIA OpenShell Sandbox, and Slack and Discord Breakout Tasks

LangChain. LangChain released deepagents-talon 0.1.0, which adds /effort controls for reasoning, an NVIDIA OpenShell sandbox provider (via a new openshell extra), and independent breakout tasks for Slack and Discord, with Discord guild mentions now answered in threads as in Slack. The release also makes background subagent capacity and checkpoint backends configurable, lets skills be pinned to scheduled jobs and reload when their files change, and stores inbound media as backend blobs. As a breaking change, Talon now requires Python 3.14. Source

2. Deep Agents Code 0.2.0 Moves to Python 3.14

LangChain. deepagents-code 0.2.0 makes Python 3.14 (>=3.14,<3.15) a hard requirement; installs with auto-upgrade enabled are migrated automatically, while others need to re-run the dcode installer or recreate their environment. The only other change listed is a fix to decode Git output as UTF-8. The preceding 0.1.88 release, published the same day, fixed async subagents so webhook configuration passes through and resolved bridge compatibility with Python 3.14. Source

3. langchain-core 1.7.0 Streams Tool-Call Arguments as Deltas

LangChain. langchain-core 1.7.0 changes streaming so tool-call arguments are emitted as tool-call-delta slices instead of block-delta snapshots of the arguments accumulated so far. Code that reads raw stream events and their block-delta args mid-stream should switch to tool-call-delta, while .tool_calls.get() and .output are unchanged. Projects that need the old snapshot behavior can pin langchain-core<1.7. Source