Sell This AI Chatbot For $10,000! (Langflow)

Step-by-step Vector Store RAG tutorial in Langflow, an open-source project for building AI agents and chatbots.

Learn how to build AI Chatbots for customer support in Langflow (e.g. Vector Store RAG), trained on your custom knowledge base and documents, leveraging an advanced LLM reasoning model, and able to analyze images.

---

Why Langflow?

If you’ve ever been frustrated by the limits (and price tags) of AI support bots like Intercom’s Fin AI Agent, you’re not alone.

My own AI support bot costs me $1,000+/month because it charges $1 per “resolution” (which is inaccurately defined)… and still manages to get many things wrong. And that number keeps growing.

Tools like Intercom’s Fin AI Agent sound amazing. They’ve come a LONG way in the past 10 years, pre-LLM era. Honestly, I loved how easy it was to plug-and-play and deploy it the first time.

However, over time I noticed limitations:

• Can’t understand screenshots

• I can’t customize the context

• Can’t select models to use

• Does not use reasoning models

• $1 per resolution (I often end up paying even if a customer’s issue is not resolved)

• external tool calls time out after 15 seconds (e.g. call my custom n8n agent)

Today, I’ll show you how to build an AI support chatbot in Langflow. It is is 100% open source and readily extensible, so you can customize it for your industry or niche, then package it up to resell... unlike other workflow builders with huge enterprise licensing fees.

In short, you’ll avoid steep enterprise licensing, own everything you build, and can fully customize it for any niche. No gatekeeping. No sketchy “$1 per resolution” charges.

---

Building the Vector Store RAG Chatbot

Step 1: Download & Install Langflow

• Go to langflow.org → Get Started for Free

• Fill out the form and choose your OS

• Launch the app

Step 2: Create a New Flow

• Click Create New Flow

• Choose the Vector Store RAG template

RAG = Retrieval Augmented Generation

Essentially, this means we’re going to train the AI chatbot on your docs to help reduce hallucinations and tailor responses to your product.

Step 3: Feed It Your Help Docs

• To keep it simple, upload a few files to start with. But in the Youtube tutorial, I also show you how to point to your website.

• On the left: choose Data → URL

• Paste the link to your help docs (e.g., help.blotato.com)

• Set Depth to 2 (scrapes all linked pages too)

• If you’re going to use the URL module, then delete the unused connector coming from the File input node.

Step 4: Connect Your Vector Store (Astra DB)

• Go to astra.datastax.com → Sign Up

• Create a database (“test” name is fine), pick your region

• Generate an application token (set it to expire in 7 days, since we’re just testing)

• Back in Langflow, paste your token, select your DB & collection

You don’t need OpenAI Embeddings blocks with AstraDB. So you can delete them to tidy up your workflow.

Step 5: Configure Your Model

• In the Retriever Flow: plug in your OpenAI API key

• Pick a model

• Tweak the prompt if you want:
  “Given the context above, answer the question as best as possible.” In my case, I added: “Return top 2 sources as URLs” so users can dive deeper)

• Click on the node to customize advanced settings — streaming, temperature, system message, etc.

Step 6: Testing

• Click Playground button in the top-right corner to test it

• Try asking questions and see how your AI chatbot performs!

• You can also deploy it on your website real quick by clicking “Share” > “Embed into Site”.

  

  

---

Closing

Overall, Langflow a nice GUI on top of Langchain. It’s intuitive if you’re already familiar with workflow builders such as Make or n8n. You can build, tweak, and own every part of your chatbot, optionally packaging it up into your own custom product.

Compared to my current Intercom Fin AI Agent, this version I whipped up in Langflow in 10 minutes already:

• uses advanced reasoning models (e.g. ChatGPT o3)

• can analyze screenshots

• doesn’t charge me $1 per resolution

However, the downsides are:

• lack of cloud hosting (you need to deploy it yourself)

• no prebuilt “escalate to human” path or UI

• no prebuilt “learn from past conversations” module

Other tools worth exploring if you're seriously looking:

• Dify.ai (open source)

• Flowise (open source)

• Botpress

• Voiceflow

The non-open source options, Botpress and Voiceflow, both allow you to choose which AI model to use, analyze screenshots, escalation-to-human paths, and “learn from past conversations” capabilities. Although their pricing model is not “$1 per resolution”, they scale up based on “credits”. Here is Voiceflow’s credit pricing calculator to compare.

---

Need More Help? 👋

1/ Free AI courses & playbooks here

2/ Free AI prompts & AI automations

3/ I built Blotato to scale myself