Clarification Request: Restore Support for Generic / Local LLMs in Core Agent Component (Target: v1.7.0)

[flefevre]

Hi Langflow team,

First, thank you for the continuous improvements and great work on the project.

I would like to raise a clarification point regarding the recent changes to the Core Agent component.
In the latest version, it seems that the component now only accepts LLMs from major cloud providers such as OpenAI, Google Gemini, and more recently IBM WatsonX.

However, the previous ability to use generic or local LLMs — especially through tools like Ollama, LiteLLM, or other self-hosted providers — appears to have been removed.
Likewise, some widely used open-source or European providers such as Mistral AI no longer seem available within the Agent interface.

This change might be intentional, a temporary simplification, or part of a broader architectural refactor — so I may be misunderstanding the intention.
To better understand and plan ahead, I would like to ask:

Could you clarify the reasoning behind this change?

• Was the removal of support for generic/local LLM integrations in the Core Agent component intentional?
• Is this limitation temporary?
• Is there a plan to restore this flexibility in the upcoming 1.7.0 release?
• Should users expect a new component or approach to handle local or custom LLM backends?

Being able to use both cloud-based and local/open-source models has always been one of Langflow’s strongest advantages.
Many users rely on Ollama, self-hosted models, or non-GAFA/M providers for privacy, cost control, or compliance reasons.

Restoring this capability in 1.7.0 would be greatly appreciated by the community.

Thank you again for your excellent work and for taking the time to provide clarification!

[andifilhohub]

I'm sorry for not getting back to you sooner. As I mentioned earlier, it is still possible to use other LLMs, including local ones, within Langflow.

It’s possible that you encountered a specific Langflow version where this option was broken and went unnoticed by the team. To help us investigate, could you let me know exactly which version you’re seeing this behavior in—where the component does not allow connecting other LLMs?

To the best of my knowledge, Langflow does not plan to drop support for local LLMs (and, frankly, I don’t see a reason why we would). If there is a limitation you’re observing, it is more likely related to a version-specific issue rather than an intentional product decision.