Langflow quickstart v1.4.0 (LTS)

This quickstart walks you through building a simple retrieval-augmented generation (RAG) flow in Langflow on Hybrid Manager. By the end you'll have a working flow that accepts a user question, retrieves relevant content from a knowledge base, and returns a grounded answer from an LLM, all running within your HM cluster.

Prerequisites

Complete the Langflow prerequisites for roles and access key setup, then make sure you also have:

• A GPU node deployed for model serving. See GPU recommendations.

• At least one model cluster deployed from the Model Library with an LLM for generation. See Creating a model cluster. The knowledge base in this quickstart uses the AIDB bert embedded model, which runs on the database CPU — no embedding model cluster is required.

• A Postgres cluster in your HM project with a table containing the content you want to query. Have the database username and password for that cluster ready — you'll need them in Step 2.

Prepare your data: create a knowledge base

Before opening Langflow, create a knowledge base in your HM Postgres cluster so the flow has content to retrieve from.

You can run this using psql or any SQL client connected to your cluster. To find the connection details, go to your cluster in the HM console and select Connect.

Run the following, replacing the table name and column with your own data:

CREATE EXTENSION IF NOT EXISTS aidb CASCADE;

SELECT aidb.create_table_knowledge_base(
    name               => 'my_knowledge_base',
    model_name         => 'bert',
    source_table       => 'public.my_content_table',
    source_data_column => 'content',
    source_data_format => 'Text'
);

SELECT aidb.bulk_embedding('my_knowledge_base');

Verify the knowledge base is ready:

SELECT * FROM aidb.kbstat;

Step 1: Accessing Langflow

1. Log into the HM console and select your project.

2. Select Launchpad in the navigation and open Langflow.

Step 2: Setting up Global Variables

EDB components read credentials from Langflow Global Variables. Set these up once and reuse them across all components in the flow.

1. In Langflow, go to Profile → Settings → Global Variables.

2. Create the following variables:

   Variable name	Value
   HM_API_KEY	Your HM machine user access key.
   HM_DB_USER	Postgres user for your HM cluster.
   HM_DB_PASSWORD	Password for the Postgres user.

See Configure Global Variables for instructions.

Step 3: Creating a blank flow

1. On the Langflow home screen, select New Flow → Blank Flow.

Step 4: Adding a chat input

1. In the Langflow canvas, select Inputs in the component panel.

2. Drag Chat Input onto the canvas. This captures the user's question.

Step 5: Adding an EDB Knowledge Base component

1. Select EDB in the component panel and drag EDB Knowledge Base onto the canvas.

2. Configure the component:

   • HM Machine User Key → select the HM_API_KEY Global Variable.
   • Hybrid Manager Project → select your project.
   • Hybrid Manager Database → select the cluster containing your knowledge base.
   • Database Connection Type → select the appropriate connection type.
   • User for the database → select the HM_DB_USER Global Variable.
   • Password for the database → select the HM_DB_PASSWORD Global Variable.
   • Database Name → select the database where the knowledge base was created.
   • Knowledge Base → select my_knowledge_base (created in the Prerequisites).

3. Connect Chat Input → EDB Knowledge Base. This wires the user's question into the Query Value field automatically.

Step 6: Adding a Prompt Template

Without a prompt template, the model receives a raw dump of retrieved chunks alongside the question. A template structures the input into a prompt the model can follow.

1. Select Prompts in the component panel and drag Prompt Template onto the canvas.

2. In the Template field, enter:

   → WrapCopy

   Answer the question using only the context below. If the answer isn't in the context, say you don't know.
   
   Context:
   {context}
   
   Question:
   {question}

3. Connect EDB Knowledge Base → Prompt Template (context variable).

4. Connect Chat Input → Prompt Template (question variable).

Step 7: Adding an EDB Model Server component

1. Select EDB in the component panel and drag EDB Model Server onto the canvas.

2. Configure the component:

   • HM Machine User Key → select the HM_API_KEY Global Variable.
   • Hybrid Manager Model Server Cluster Instance → select the LLM model cluster deployed in the prerequisites.

3. Connect Prompt Template → EDB Model Server.

Step 8: Adding a Chat Output

1. Select Outputs in the component panel and drag Chat Output onto the canvas.

2. Connect EDB Model Server → Chat Output.

Step 9: Running and testing the flow

1. Select Playground in the top toolbar to open the interactive test panel.

2. Type a question that relates to your source data and press Enter.

3. Verify the response is grounded in your content — you should see the EDB Knowledge Base retrieving relevant rows, the Prompt Template formatting the context and question, and the EDB Model Server generating a contextualised answer.

4. Check that all components show green status indicators after running.

If any component shows an error, select it and review the configuration. Common issues:

• Can't select a project in the EDB Knowledge Base component — the portal domain is defaulting to the API endpoint. Select Controls on the component and set Hybrid Manager URL to http://istio-internalgateway.istio-system.svc.cluster.local:7080.

• Incorrect or expired access key — verify the HM_API_KEY Global Variable holds a valid machine user key.

• Mismatched database connection — confirm the database user, password, and database name match what's configured in the cluster.

• Model cluster not active — check the cluster status in Agent Factory → Model Server Clusters.

For further help, see Troubleshooting.

Next steps

• Add more components — extend the flow with an EDB Database component to query structured data alongside the knowledge base, or an EDB Vector Index component to manage vector indexes. See EDB components for the full set, including EDB Airman MCP, EDB Platform, and EDB Embedded Models.

• Deploy the flow — publish it as a bundle and deploy it as a managed, callable service with its own endpoints, project-level isolation, and independent scaling. See Flow deployment.

• Monitor the flow's dependencies — set up Prometheus alert rules and Grafana dashboards for the model cluster and Langflow pods. See Observability.