⚠️ Recommending LangGraph Platform for new projects

[eyurtsev]

We have recently announced LangGraph Platform, a significantly enhanced solution for deploying agentic applications at scale.

We recommend using LangGraph Platform rather than LangServe for new projects.

LangGraph Platform incorporates key design patterns and capabilities essential for production-level deployment of large language model (LLM) applications.

In contrast to LangServe, LangGraph Platform provides comprehensive, out-of-the-box support for persistence, memory, double-texting handling, human-in-the-loop workflows, cron job scheduling, webhooks, high-load management, advanced streaming, support for long-running tasks, background task processing, and much more.

The LangGraph Platform ecosystem includes the following components:

• LangGraph Server: Provides an Assistants API for LLM applications (graphs) built with LangGraph. Available in both Python and JavaScript/TypeScript.
• LangGraph Studio: A specialized IDE for real-time visualization, debugging, and interaction via a graphical interface. Available as a web application or macOS desktop app, it's a substantial improvement over LangServe's playground.
• SDK: Enables programmatic interaction with the server, available in Python and JavaScript/TypeScript.
• RemoteGraph: Allows interaction with a remote graph as if it were running locally, serving as LangGraph's equivalent to LangServe's RemoteRunnable. Available in both Python and JavaScript/TypeScript.

If you're interested in migrating your LangServe code to LangGraph Platform please the LangGraph Platform Migration Guide for more information.

We will continue to accept bug fixes for LangServe from the community; however, we will not be accepting new feature contributions.

[ewiner]

My understanding is that LangServe and LangChain are open-source, while LangGraph Server is a commercial product with closed-source dependencies and limitations on free usage (according to the deployment options page). Is that correct?

[eyurtsev]

Yes, it's a commercial offering. The free tier option allows up to 1 million nodes executions and has an option for self hosted deployment (https://github.com/langchain-ai/langgraph/blob/main/docs/docs/concepts/deployment_options.md#self-hosted-lite).

[eyurtsev]

@prise6, @madoe001, @timvw, @aaronvenezia, @laithalsaadoon, @heidar, @teeppp, @cycleuser0x1 would appreciate any feedback on the 👎 .

Is there something that you'd like LangGraph to do that it doesn't but LangServe does?

[chai3]

This news made me sad.
My chain doesn't need LangGraph's memory or checkpoints.
LangGraph requires Postgresql and Radis, which makes the infrastructure cumbersome.
I want to host my Chain with simple infrastructure and add it to the existing FastAPI routes (langserve.add_rouets).
I run LangServe with:
AWS Lambda + web adapter
AWS AppRunner

[laithalsaadoon]

TBH the news is disappointing because all the **new** serving features are behind a Langsmith key 🤷

…

On Sat, Dec 28, 2024, 3:04 AM chai3 ***@***.***> wrote: This news made me sad. My chain doesn't need LangGraph's memory or checkpoints. LangGraph requires Postgresql and Radis, which makes the infrastructure cumbersome. I want to host my Chain with simple infrastructure and add it to the existing FastAPI routes (langserve.add_rouets). I run LangServe with: AWS Lambda + web adapter AWS AppRunner — Reply to this email directly, view it on GitHub <#791 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACI4QLQL2PTET3NXKU2ECCD2H2AVLAVCNFSM6AAAAABSAAQJHSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKNRUGI4TOMZVGQ> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[aaronvenezia]

Feels like a rug pull to me. You built a tool for deployment just to abandon it for a commercial one with your now captive developers that have invested time in building with your tool. I'm fine with commercial products, but (just my opinion) the new offering shouldn't have a limitation when it comes to self deployment. That would make the transition a lot easier to swallow. I understand that's what happens sometimes, it's just a little disappointing.

That said, you guys build a lot of great tools and I very much appreciate all that you provide in the AI framework space. Thank you for all you do.