One of the most common frustrations when building complex AI agents in n8n is hitting that 'Prompt too long' error or seeing the model lose track of instructions as the workflow grows. The secret to scaling isn't a bigger model—it's **MODULARIZATION**. Instead of building one massive workflow that tries to do everything (and stuffs the entire context into one prompt), I've been using a 'Multi-Agent Factory' approach: 1. **The Orchestrator:** A main workflow that decides which task needs to be done. 2. **The Workers:** Separate workflows for specific tasks (e.g., Data Extraction, Analysis, Formatting) called via the 'Execute Workflow' node. 3. **The Memory:** Using a centralized database or Notion CRM to store intermediate states so each sub-workflow only gets the context it actually needs. This keeps your prompts clean, your executions fast, and your debugging way easier. How are you guys handling context limits in your production builds? Are you using vector DBs or just aggressive chunking? Let's discuss! 👇

hi. im interested in digitizing all the Epstein files into vector format and then reference training it with a LLM model. After that I want to convert and tag the data into markdown files that would be easily searchable via visual mind maps and context search. this is a volunteer type project, meaning no money involved but this would look great on a resume, portfolio and online as we discover nuggets of information that others have missed and post that information online. the tools we plan to use are n8n, obsidian, qdrant, docling. if you are interested in participating dm and tell me about your relevant experience and why you are interested. thx