Hello, ComfyUI community! 👋

I'm thrilled to announce Workflow Architect v0.4 — a major update that makes your AI assistant even smarter at discovering and using workflows from across your entire ComfyUI ecosystem!

This release brings Blueprint scanning, automatic discovery of example workflows from custom nodes, and deeper RAG integration. The system now learns from every JSON template it can find — whether it's a built-in blueprint, a user template, or an example provided by a custom node author.

It's still a beta, and your feedback is invaluable. Let's dive into what's new! 🎉

🔥 What's New in v0.4

📋 Blueprint Scanner
ComfyUI Frontend stores blueprints in ComfyUI/blueprints/. Workflow Architect now scans this directory (alongside web/templates/, templates/, and user/templates/) to discover even more pre-built workflow structures. Every found blueprint is added to the RAG index for better generation quality.

🧩 Custom Node Workflow Discovery
This is a game-changer! The scanner now recursively looks inside your custom_nodes/ folder and finds JSON workflows stored in:

• custom_nodes/*/example_workflows/

• custom_nodes/*/workflows/

• custom_nodes/*/examples/

Any JSON file found there is automatically added to the knowledge base with source="custom_node" and the node's name attached. This means:

• If a custom node author provides example workflows, they're now part of your AI's knowledge.

• The system learns from real-world usage patterns across all your installed extensions.

• You can drop your own workflows into these folders and they'll be discovered instantly!

🧠 RAG Index with Source Attribution
Every document in the RAG index now carries metadata about its origin:

• source: "comfyui_default" — built-in ComfyUI templates and blueprints.

• source: "custom_node" — workflows found inside custom nodes, with custom_node_name attached.

This helps the system understand which templates are authoritative and which come from community examples.

🔍 Expanded Template Scanner
The scanner now checks:

1. ComfyUI/web/templates/ (legacy interface templates)

2. ComfyUI/blueprints/ (Frontend blueprints) ← NEW

3. ComfyUI/templates/

4. ComfyUI/user/templates/ (user-saved templates)

5. custom_nodes/*/{example_workflows,workflows,examples}/ ← NEW

⚙️ Smart Filtering
Workflow discovery now intelligently filters base templates (only shown if all required node types are installed) and auto-generates showcase workflows for additional installed nodes that aren't in the base set.

📊 Enhanced Stats Endpoint
The /architect/stats endpoint now reports template counts from all sources, giving you a clear picture of what the system has discovered.

🔑 Important: Embedding Model Requirements

For the RAG (Retrieval-Augmented Generation) system to work properly, you need a separate embedding model for document indexing and similarity search. Chat models like llama3.1 or gpt-4 do NOT support embeddings!

Default embedding models per backend:

BackendDefault Embedding ModelNotesOllamanomic-embed-textYou need to pull it: ollama pull nomic-embed-textOpenAItext-embedding-ada-002Available by default with API accessllama.cppUses the same model as chatRequires a model with embedding support (e.g., nomic-embed-text-v1.5)

⚠️ Important: If you're using Ollama (the default), make sure you have the embedding model pulled:

bash

ollama pull nomic-embed-text

Without this, the RAG engine will fall back to keyword search (less accurate) or fail to index documents properly. The extension will warn you in the logs if embeddings aren't available.

🐞 Beta Reminder & Call for Help

This is still an experimental release. The new scanner logic may occasionally pick up invalid JSON files or misidentify templates. Your testing is crucial!

Please:

1. Test the scanner with your setup — does it discover your custom node examples?

2. If you notice incorrect templates being loaded, or if the scanner misses obvious ones, please report it!

3. Share logs (comfyui.log) and any error messages — they're super helpful for debugging.

📥 How to Install / Update

1. If you're updating, remove the old version from custom_nodes/ first.

2. Copy the new comfyui_workflow_architect folder into custom_nodes/.

3. Install dependencies (if not already installed):

   bash

   pip install -r requirements.txt

4. For Ollama users: Pull the embedding model:

   bash

   ollama pull nomic-embed-text

5. Restart ComfyUI.

That's it! The scanner will run automatically on startup and seed the RAG index with everything it finds.

Version: 0.4-beta
Author: InsanE_GeN (Civitai)

Thank you for your continued support and testing! Your feedback directly shapes the future of Workflow Architect. Let's make ComfyUI even more powerful, together! 💪🎨

P.S. If you're a custom node author — consider adding an example_workflows/ folder to your node with sample JSONs. Workflow Architect will automatically discover them and make your node even more accessible to users! 🙌