New video: Build a streaming @LangChainAI agent in @nextjs using useStream + memory 🚀 You’ll learn: - stream AI replies into your UI with useStream - Minimal API route serving SSE - Add conversation memory via thread id + checkpointer 🎥 Watch now: https://t.co/1ysdHDOwXM

🫎LangSmith in the wild Cool to see developers sharing public links of their traces when they share new projects! Nothing better than seeing exactly whats going on under the hood of the agent, great way to understand an agent

You can use @AnthropicAI's structured output for Sonnet or Haiku 4.5 with @LangChainAI by importing and using the `providerStrategy` .. that's it, zero additional config needed! 🙌 We will make it the default output strategy once it comes out of beta.

🌶️ Hot take: Most “AI agents” today aren’t production-ready — not because of the LLM…but because one flaky tool brings the whole system down 🧨 In the new video, I show how LangChainJS’s Tool Retry Middleware gives agents real resilience: 🔁 automatic retries ⏱️ exponential

If you use #AI to build workflows 🤔 you are doing it wrong! Introducing @LangChainAI Agent Builder 🚀 now in public beta! I built a research agent in 5 minutes while other senior engineers are still arguing about which orchestration pattern to use. 👉

🧠 New from LangChain: Automatic Summarization Middleware Build agents that handle long, multi-turn conversations without blowing through your context window. Older messages get compressed, while all of the details in recent history are preserved. You can configure

LLM vendors keep shipping bigger context windows… but nobody talks about the actual problem: Tool-heavy agents drown in their own tool results 🤯 @AnthropicAI fixed this with context editing. We made the idea model-agnostic in @LangChainAI 🚀✨ 🎥 https://t.co/8xGgEAekHi

Voice agents shouldn't require vendor lock-in. If you're building voice with @LangChainAI, this might save you weeks. GitHub: https://t.co/NpBczl7l9x Live Demo: https://t.co/7jQa43SwmP Let me know what providers you'd want added! 👇

🥪 Try the "Voice Sandwich" demo: 1. Go to https://t.co/7jQa43SwmP 2. Pick your STT (AssemblyAI or OpenAI) 3. Pick your TTS (ElevenLabs, Hume, or OpenAI) 4. Order a sandwich with your voice The voice agent will take your order, manage the cart, and hang up when done. 🧵 5/6 👇

The secret? Everything is a TransformStream: Audio → [beforeSTT] → STT → [afterSTT] → Agent → [beforeTTS] → TTS → [afterTTS] → Audio Middleware hooks at every stage. Just like @LangChainAI createMiddleware(), but for voice. 🚀 🧵 4/6 👇

create-voice-agent extends LangChain's createAgent() with voice: ✅ Hot-swap providers at runtime ✅ Streaming TransformStream pipeline ✅ Built-in barge-in support ✅ "Let me think..." filler middleware ✅ Works with your existing @LangChainAI tools One function. Full voice

The voice agent landscape is fragmented: ❌ Tied to one STT/TTS provider ❌ 500ms+ latency breaks conversation flow ❌ User interrupts → system crashes ❌ Building pipelines from scratch takes weeks ❌ no tracing or evals Developers are struggling. 🧵 2/6 👇

🎙️ Building voice agents is hard. Provider lock-in. Latency hell. Barge-in nightmares. I built a harness on @LangChainAI that solves all of it. Switch between @AssemblyAI, @OpenAI, @elevenlabsio, @hume_ai with a DROPDOWN 🤯 Try it live: https://t.co/7jQa43SwmP Thread 🧵👇

Want to learn about these CLI’s everyone’s talking about? @LangChainAI’s OSS Deep Agents CLI is a fun way to start Built on LangChain’s OSS Deep Agents library, the CLI will introduce you to subagents, task lists, memory, *skills* + more uv pip install deepagents-cli

💬 Join our Community Jam Session on Dec 15 (Dec 16 for some APAC zones)! Share your feedback on LangChain 1.0 & 1.1, what’s working, what’s not, and what you love. Your input shapes LangChain’s future! 🚀 👉 RSVP: https://t.co/Uuu6z6m0FQ

Your agent gets slower, sloppier, or “dumber” the longer the thread runs? That’s context overload! 🤯 In my new video I break down how @LangChainAI's Summarization Middleware keeps long-running agents sharp by automatically compressing history without losing important tool I/O.

So cool to see all these @LangChainAI agents doing amazing work 👏🚀

AI agents are amazing�� until they start acting like unhinged interns🤪 Without guardrails, they can overshoot rate limits 📈, blow through API quotas 💸, or trigger endless tool call loops 🔁. @LangChainAI's Tool Call Limit Middleware gives you precise control: per run, per

The team @LangChainAI is working on making streaming messages to the frontend much easier as currently implemented in LangChat 🚀 so keep an eye on updates soon 😊 I am also planning on publishing a version for this in @vuejs , @angular and @sveltejs

If you want to experiment with different #AI agent behavioral patterns, I've build https://t.co/Acj3w1jCDq to simulate features like model/tool limit, retries, HITL or different context engineering approaches using @LangChainAI 👀 it's a simple @nextjs and the code is available

Day 0 support of @Google Gemini 3 in @LangChainAI 🔥🚀