Pinned
Sydney Runkle
676 posts
💁♂️ Introducing human-in-the-loop Middleware! 💵 Tool calls can be risky and expensive! For certain tool calls you may want to get user feedback to approve, deny, or modify them before execution. Our new middleware provides an easy off-the-shelf way to build this into your
- 🚀 LangGraph v0.6.0 is here! This release brings: ✨ A new context API for cleaner, type-safe runtime dependency injection 🔀 Dynamic model & tool selection for create_react_agent 🛡️ Enhanced type safety & autocomplete for graph building and invocation 🏗️ Durability mode for
- Context engineering is the number one way to get your agents to perform well. But it's hard to get right!! DeepAgents handles some of the hardest (but most important parts) for you under the hood by: 🧠 compressing context w/ summarization 🗄️ offloading context w/ the fileToday we're releasing 0.2 of deep agents 📁The main addition here is a "backend" abstraction, which allows you to swap out the filesystem that deep agents use. Can be local filesystem, a database, remote vm, anything Agent Harnesses like deep agents are becoming more important
- 📖Just published a new page on context engineering for agents with @langchain. 🍰Icing on the cake w/ our new docs! docs.langchain.com/oss/python/lan…
- @LangChain V1 middleware explained in < 30 seconds: 🔍 before_agent — Load files, validate input ✂️ before_model — Summarize convos, trim messages 🔄 wrap_model_call — Dynamic prompts, model, tools 🛠️ wrap_tool_call — Tool retries, error handling 🤵after_model — Human in the
- MCP was missing key ingredient for agentic systems - until today! A critical part of multi-agent workflows is human in the loop (HITL) functionality. Every agent in a system should be able to interact with a human supervisor. With "elicitation" (announced today), MCP has takenAfter a long discussion, we landed a first version of Elicitation to the MCP draft specification. Lots more to work on and to evolve. Take a look modelcontextprotocol.io/specification/…
- 📣 I'm giving a talk tomorrow at @_odsc west on context engineering w/ LangChain + LangGraph! 📄 Perhaps the biggest news with the LC + LG V1 releases is the rollout of our new, unified docs site. Check out our guide on context engineering w/ agents as a preview!
- 🚀LangGraph v0.5 is out! Check out the new features and improvements in our release notes!
- 🔁🛠️Dynamic Tools in LangGraph Agent Effective agents often need access to different tools throughout a single run. For example, you might want to force the agent to use a smaller subset of tools at the start. Docs: langchain-ai.github.io/langgraph/how-…
- 🎬 Starting TODAY: our new LangChain middleware series! One deep-dive video per day covering @langchain's prebuilt middleware options. 👤 Day 1: Human-in-the-Loop Middleware Use this to require human review of risky tool calls before execution!
- check out our new coding agent CLI, built on top of (in this order): * deepagents (agent harness) * langchain v1 (agent framework) * langgraph v1 (agent runtime)💻Deep Agents CLI Using the deepagents package, we built a simple coding CLI as an example of a coding application you could build on top of deepagents We added in a concept of memory so that it would remember instructions and guidance over time Blog: blog.langchain.com/introducing-de…
- 🔬 OpenAI just released new deep research models. 🤖 You can build agents with those models TODAY in langgraph!
- 💰 Not every task needs your most expensive model! 🔀 Use dynamic model selection in LangGraph to auto-escalate from cheap to powerful models based on task complexity!
