Description

Langflow already ships with vLLM and vLLM Embeddings components (src/lfx/src/lfx/components/vllm/), and the vLLM icon is already registered in the frontend icon system. However, vLLM is not included in the Model Providers list (Settings → Model Providers), which means users cannot configure a vLLM server centrally and must manually enter the API base URL in every vLLM component instance.

Current Behavior

• Model Providers list includes: OpenAI, Anthropic, Google Generative AI, Ollama, Groq, Azure OpenAI, IBM WatsonX
• vLLM components exist but require manual per-component configuration
• Users cannot select vLLM models from the Language Model / Agent component dropdown

Expected Behavior

• vLLM should appear in Settings → Model Providers alongside Ollama (both are self-hosted, OpenAI-compatible servers)
• Users configure VLLM_API_BASE (required) and optionally VLLM_API_KEY once in settings
• vLLM models are dynamically discovered from the server (like Ollama) and appear in the Language Model / Agent dropdowns
• No new LangChain dependency needed — vLLM uses the OpenAI-compatible API (ChatOpenAI)

Why vLLM?

vLLM is one of the most popular open-source LLM serving frameworks, widely used in enterprise and research environments for self-hosted model inference. It provides an OpenAI-compatible API endpoint, making integration straightforward. Given that Langflow already has vLLM components and icons, adding it to Model Providers is a natural and low-effort improvement.

Implementation Notes

• Add "vLLM" to LIVE_MODEL_PROVIDERS (dynamic model discovery, like Ollama)
• Add vLLM entry to MODEL_PROVIDER_METADATA with VLLM_API_BASE and VLLM_API_KEY variables
• Add vLLM server validation in validate_model_provider_key() — simple GET /models health check
• vLLM icon (VllmIcon) is already registered in eagerIconImports.ts
• vLLM uses ChatOpenAI class (already imported), so no new dependency

Files to Modify

1. src/lfx/src/lfx/base/models/model_metadata.py — Add to LIVE_MODEL_PROVIDERS and MODEL_PROVIDER_METADATA
2. src/lfx/src/lfx/base/models/unified_models/credentials.py — Add vLLM validation logic