LangChain OSS

Ship great agents fast with our open source frameworks – LangChain, LangGraph, and Deep Agents. Maintained by @LangChain.

[Sydney Runkle @sydneyrunkle]

gemini decisions API launched today!

this API will feel very familiar if you've used jev or other decision models. gemini decisions is much faster and cheaper than an LLM, and outputs are structured + reliable :)

we have day 0 support in langchain: https://t.co/UU60VcPTqi

[LangChain OSS @LangChain_OSS]

docs: https://t.co/cE7AxDARGO

[LangChain OSS @LangChain_OSS]

LangChain has day zero support for Google's new Decisions API!

Gemini + TypeSafe classifiers share a base class, so middleware takes either.

[LangChain OSS @LangChain_OSS]

LangChain Community Spotlight: LangChain Explained by our Poland Ambassador 🇵🇱

Simon Budziak (@simon_budziak), our Poland Ambassador, covers messages in under a minute for episode 8 of LangChain Explained. He shows that everything a LangChain agent knows lives in one list of messages, and the model rereads all of it on every turn. His advice is to put your effort into deciding what stays in that list.

📺 Watch the episode: https://t.co/5eOhakcqUJ

[LangChain OSS @LangChain_OSS]

LangChain Community Spotlight: Querit Search Tools 🔍

The Querit team (@QueritAi) built LangChain tools for their search API. WebSearchTool returns up to 50 results for a query, and WebContentsTool fetches the full page behind up to 10 URLs per call. A page that fails to crawl comes back marked as failed instead of breaking the whole call.

🔗 GitHub: https://t.co/XvtMVrt7vZ

[LangChain OSS @LangChain_OSS]

LangChain Community Spotlight: Enterprise RAG on LangChain 🔎📚

When a RAG answer comes back wrong, Kacper Klusek of Soba Labs reads what retrieval returned before blaming the model. His guide builds the retrieval stack on LangChain step by step, and in Anthropic's published evaluation, the chain he recommends cut the top-20 retrieval failure rate from 5.7% to 1.9%.

📖 Read the full guide: https://t.co/Bi5HJATM7Z

[LangChain OSS @LangChain_OSS]

Async subagents keep your agent responsive to you while background workers handle the long tasks.

https://t.co/oLJoHxPwpS

[LangChain OSS @LangChain_OSS]

LangChain Community Spotlight: Which User Is Costing You Money? 💸

Your agent bill went up and you want to know which user did it. Soba Labs shows how to find out by passing a user_id in each run's metadata, then grouping LangSmith's dashboard by user to see cost per customer. They also found that behind an OpenAI proxy, streamed token usage can read zero, so check one trace by hand before you trust the split.

📺 Watch the episode: https://t.co/LWXP7X87JU

[LangChain OSS @LangChain_OSS]

LangChain Community Spotlight: 50 Advanced Generative AI Concepts 🧠

Gustaf, a senior AI engineer, explains 50 GenAI concepts and their trade-offs. In the section on agent skills, he points out that an agent picks what to load from each skill's short description, so a skill with a weak description might never be found. He suggests using LangSmith to check whether your agent actually loads the skills it's supposed to.

📺 Watch the full breakdown: https://t.co/Xu3aKTryv5

[LangChain OSS @LangChain_OSS]

LangChain Community Spotlight: How Deep Agents Run the Porting Loop 🎮🤖

Bart Ludera (CEO, Harnessloop) writes up a production game port where an agent built on LangChain's Deep Agents took over the build-and-test loop a programmer used to drive by hand. It reruns the same scenario before calling a defect fixed, and the programmer now steers several of these agents at once.

📖 Read the full case study: https://t.co/U3IBVnmmPz

[LangChain OSS @LangChain_OSS]

Async subagents keep your agent responsive to you while background workers handle the long tasks.

Try them out in Deep Agents today!

[Sydney Runkle @sydneyrunkle]

https://t.co/9loyF9vcBa

[Sydney Runkle @sydneyrunkle]

if you give your agent a browser, it can do work on the web, even on sites w/ no agent native APIs!

in this demo my agent books dinner reservations for me. i only had to log into the reservation site once; it saved my login and acts on my behalf for every new session!

[LangChain OSS @LangChain_OSS]

Give your agents the web w/ browser use in deepagents!

[Sydney Runkle @sydneyrunkle]

agentic browser requests jumped ~5x in september alone... browser use is becoming a key primitive for agents!

icymi, we shipped it in (managed) deep agents last week

now your agent can do anything interactive on the web!

https://t.co/nnskeujiKJ

[LangChain OSS @LangChain_OSS]

LangChain Community Spotlight: 50 Advanced GenAI Concepts 🧠

Gustaf, an AI Engineer, walks through 50 concepts that come up in AI engineer interviews, from KV cache and GraphRAG to agent harnesses and LLM-as-a-judge evals. The agent section covers using LangSmith to check whether your agent actually loads its skills.

📺 Watch the full breakdown: https://t.co/Xu3aKTryv5

[Sydney Runkle @sydneyrunkle]

i feel like we need a new message type! injected like a human message but not from a human. like:

* async subagent returns (or a2a stuff)
* browser handed back to user
* summarization event
* skill injection

a notification type? has anyone thought deeply about this?

[Sydney Runkle @sydneyrunkle]

giving a talk tomorrow on "building agents that work like coding agents"

at @LangChain we help folks build all types of agents, but a lot of them "work like" coding agents in terms of harness design

https://t.co/YVkszMz3Ha

[Sydney Runkle @sydneyrunkle]

joined @TalkPython w/ @mkennedy, @reuvenmlerner, @calvinhp, and pat decker to talk about "the state of python in 2026"

so much of it came back to development w/ agents! harnesses, observability, and why ~90% of my coding happens in slack now

https://t.co/Rg5DSZkrZl

[Sydney Runkle @sydneyrunkle]

harness² from @GoogleResearch offers techniques for RSI on the harness, with a fixed model

they call it "harnessing", but you might already be familiar w/ this process! it's also called "hill climbing" or "continual learning"

more in 🧵 !

[Sydney Runkle @sydneyrunkle]

sourcing ideas for the next deepagents feature on my morning commute

thanks for the inspo @browserbase

[Sydney Runkle @sydneyrunkle]

we just revamped human in the loop middleware in langchain!

interrupts on tool calls now ship w/ typed schemas, so the human's response is validated before it goes back to the agent

you can also use the schema to render nicer approval forms in your ui

https://t.co/OYxg3pSAjA