After this lesson, you will be able to: Set up LangFlow, build a RAG pipeline visually, and run a multi-turn conversational agent, all from the canvas.
LangFlow is the visual front-end to LangChain. Every node is a real LangChain component, every edge is a Python data flow. This lesson installs LangFlow, builds a document-grounded conversational agent, and walks through what each LangChain component is doing under the hood.
1. Make sure you have Python 3.10+ (`python --version`).
2. Install: `pip install langflow` (or `uv pip install langflow`).
3. Run: `langflow run`. It serves at http://localhost:7860.
4. Sign up locally. You'll see the templates gallery.
Left panel: components organized by type (Models, Memories, Tools, Vector Stores, Loaders). Center: the canvas. Right: an inline chat to test. Drag a component → connect handles by type, handles are color-coded and only compatible types snap together.
1. From templates, pick 'Vector Store RAG'. Or build from scratch:
2. Drag 'File' loader. Upload a PDF (any 5–20 page doc).
3. Drag 'Recursive Character Text Splitter'. Connect File output → Splitter input.
4. Drag 'OpenAI Embeddings' (or 'HuggingFace' for free). Add API key as env var.
5. Drag 'Astra DB' or 'Chroma' vector store. Connect Splitter + Embeddings into it.
6. Drag 'Chat Input' → 'Prompt' → 'OpenAI / Claude' LLM → 'Chat Output'. Connect the vector store output as 'context' into the prompt.
7. Hit Build, then test in the inline chat: ask a question only your PDF can answer.
1. Drag a 'ConversationBufferMemory' node.
2. Connect it as 'memory' input to the prompt.
3. Now it remembers within the conversation. Ask follow-up questions like 'expand on point 2', it will use the prior turn as context.
Sign in and purchase access to unlock this lesson.