LangChain OSS
@LangChain_OSS
Ship great agents fast with our open source frameworks – LangChain, LangGraph, and Deep Agents. Maintained by @LangChain.
6 Following 4.2K Followers
building an agent (model + harness) is about 2 things:
1. picking the right model for the job. this means finding the sweet spot on the cost/intelligence curve.
2. building a harness that's fit to the agent's task(s). this means the harness can get the right context to the model at any step.
here's a guide on how to build a domain specific harness:
Show more
The biggest challenge facing an agent harness is context engineering, or what information an agent gets, and when.
The longer the task, the harder it is to keep the context window from overloading.
Deep Agents ships with this built in: filesystems, subagents, and skills, already wired up for context management.
Show more
we just changed how deepagents read files!
our evals show that this new format reduces edit_file errors by 15% AND total input token usage by 10%
the little things really add up w/ agents; we're hard at work on these small but mighty optimizations at @LangChain
Show more
we just added a model router to our internal coding agent! this is an experiment in:
1. reducing coding agent costs
2. developing an effective task-specific router
hopefully this router design (sans domain-specific routing information) will be generalizable across agents!
is anyone having success w/ routers in the harness? if so, how did you design yours? how are you improving it?
Show more
subagents are great for parallelizing work and delegation to specialized agents, but it's hard to figure out exactly what context to pass to and return from subagents!
this guide details the two input context modes we now support
1. isolated -- subagent gets a completely new prompt
2. forked -- starts exactly where the main agent left off, with a copy of the old message history
as well as when you might want to use each!
Show more
MCP usage is growing... and growing fast!
check out our latest blog on our revamped support for MCP and the new, stateless spec!
RT @sydneyrunkle: we now support the new stateless MCP spec in @LangChain!
this revamp includes moving MCP support into the main langchain…
you can now build deepagents on top of @MongoDB Atlas!!
this backend, built on top of an underlying object store, supports more robust search including MongoDB search, vector search, and hybrid search!
check out the blog to learn more!
Show more
@LangChain has day 0 support for Gemini 3.8 flash!
you can now build @LangChain agents w/ agentic video understanding!
google is consistently at the forefront of multimodal processing; this is the video analog of "agentic vision"
token usage + cost is also way down w/ this new approach!
docs:
Show more
we are rolling support for the new MCP spec in langchain and need feedback on our new API!
with this new pattern you connect langchain agents to multiple MCP servers, hook up oauth, use elicitation w/ langchain interrupts!
we're building on top of FastMCP which is the easiest way to get started building MCP clients and servers
Show more
new MCP spec support is coming soon in @LangChain (and deepagents)
try it out and let us know what you think!
we're adding support for the new MCP spec in @LangChain open source!
the new API is built on top of FastMCP, so you can take advantage of their excellent devx for building MCP servers and clients.
an early version is available in langchain==1.4.0a2; let us know what you think!
Show more
RT @sydneyrunkle: last night i spent some time building a browser agent with stagehand from @browserbase and deepagents from @LangChain
i…
What's new in LangChain? 🚀
🧯 Standard exceptions: chat models now raise standard exception types across providers, so you can distinguish retryable errors (timeouts, rate limits, etc.) in a consistent way. Fully backward compatible.
🧩 Agent middleware: custom token_counter in ContextEditingMiddleware, plus non-retryable exceptions now skip retries in ModelRetryMiddleware (
🔥 Fireworks document reranking: a new reranker integration (
📊 Token & usage accounting: bug fixes for xAI and DeepSeek (
⚡ Portability & perf: lazy transformers import, and tighter grep scope for Anthropic (
This is your harness! Thanks to everyone who made it better. 🙏
Show more
RT @sydneyrunkle: you can now make your deepagent a browser agent w/ stagehand by @browserbase!
connect tools so your agent can click, typ…
working on support for MCP v2 for langchain/deepagents this week!
curious -- is anyone using prompts and resources? 99% of what i hear is about tools
i've heard a some buzz about MCP apps recently too, are there other features i should dig into?
Show more
we're hiring open source devs @LangChain
looking for people who are building at the frontier of agents and are excited to quickly develop ownership
apply:
What's new in LangChain? 🚀
🔌 Support for OpenAI's 3.0 SDK (using httpx2)
✨ Support for gemini-3.7-flash
Plus a wave of core reliability fixes from external contributors:
🛠️ Tool calling & structured output: clearer errors, respect for pydantic aliases when validating tool inputs (
📊 Usage metadata fixes: token usage callback bugs, and cost metadata when streaming OpenRouter (
🧱 Content & prompt hardening: guard malformed Anthropic content blocks, and preserve non-str/non-dict items in prompt templates for content blocks (
Huge thanks to everyone who contributed! 🙏
Show more
New langchain python release! 🚀 Highlights:
• Internal middleware model calls no longer leak into messages / event stream ( reported by
• SummarizationMiddleware retries failed summaries (reported by + faster token counting for tool schemas
• Batch of langchain-anthropic bug fixes (
• OpenRouter preserves provider info in response metadata (
This release also includes cross-provider standardization of reasoning effort, including supported values per-model in model.profile:
Huge thanks to our external contributors for helping to shape and improve our harness!
Show more
an RLM harness is a harness that supports "recursive language models"
here's how you can add RLMs to deepagents!
"RLM harness" is a buzzy term right now, what does it mean?
RLM = recursive language model. In practice this means a supervisor model can break down a task into smaller pieces and recursively call itself on those smaller tasks.
Here's how you can use RLMs w/ deepagents!
Show more
