How models are consumed in Langflow
In Hybrid Manager, Langflow connects to AI models through two EDB components: EDB Model Server and EDB Embeddings. Both components communicate with model endpoints that are governed and served by Hybrid Manager — no external API calls are made by default.
- EDB Model Server — sends prompts to a deployed LLM and returns generated responses. Use this for conversational steps, summarisation, and reasoning within a flow.
- EDB Embeddings — converts text into vector representations using a deployed embedding model. Use this when building retrieval steps or populating knowledge bases from within a flow.
Both components require a model cluster to be deployed in your HM project before use. See Creating a model cluster for setup steps.
Using the EDB Model Server component
In Langflow, select EDB in the left-side component panel.
Drag EDB Model Server onto the canvas.
In the component, set HM Machine User Key to your access key Global Variable.
Under Hybrid Manager Model Server Cluster Instance, select the deployed model cluster you want to use.
Connect the component's input to your flow's upstream step (for example, a Chat Input or a retrieval result).
Connect the output to a Chat Output or the next step in the flow.
Select the Play icon to run the component. Select Model Response in the lower-right corner to inspect the output.
Models available in the dropdown are those deployed from the Model Library into your HM project. Hybrid Manager manages their internal endpoints — you don't need to configure URLs manually.
Using the EDB Embeddings component
In Langflow, select EDB in the left-side component panel.
Drag EDB Embeddings onto the canvas.
Set HM Machine User Key to your access key Global Variable.
Under Hybrid Manager Model Server Cluster Instance, select the embedding model cluster.
Connect the component's Embedding Model output to a downstream Langflow vector store or retriever component that expects an embeddings model.
EDB Embeddings produces a LangChain Embeddings model object, not a vector directly. Downstream components use it to encode text at query or indexing time. It has no text input of its own.
Knowledge bases and retrieval
To ground model responses in your organisation's data, combine the EDB Model Server with the EDB Knowledge Base component:
Set up a knowledge base in your HM project using AIDB SQL or Pipeline Designer.
Add an EDB Knowledge Base component to the flow and configure it with your production project, cluster, and knowledge base name.
Add a Prompt Template component between the EDB Knowledge Base and the EDB Model Server. Connect the EDB Knowledge Base output to the template's context variable, and the user's query to its question variable. Connect the template output to the EDB Model Server input.
This pattern implements retrieval-augmented generation (RAG) entirely within your HM cluster — source documents, embeddings, and model calls all stay inside your environment.
See EDB Knowledge Base component for the full configuration walkthrough.
Environment and service discovery
When Langflow runs inside Hybrid Manager:
Model endpoints are automatically discovered from the HM Model Library — no external URLs to configure.
Authentication is handled through the HM Machine User Key you set as a Global Variable.
All traffic between Langflow, model server pods, and Postgres clusters stays within the project namespace.
If a component can't resolve the project because the portal domain defaults to the API endpoint, select Controls on the component and set Hybrid Manager URL to http://istio-internalgateway.istio-system.svc.cluster.local:7080.