Hi Cole,
Great video on Pedantic AI and LangGraph! I’m working on a new project where I build a series of agents to turn the “average Joe” blue-collar worker into a rockstar at work/jobsite. I want to get a framework like Pedantic AI/LangGraph, or N8N, or Boltdiy in place so I can spend most of my time on the jobsite collecting the relevant disparate data sources and then plug and play to build a custom ai agent. For example, one idea is a “framer’s assistant” on a construction site. “Double check measurements - 7 ¾ minus 5/8=7 1/8”, “Build me a lumber take-off order”, “order us some uber eats to jobsite”, set reminders/to-dos for next crew or following day, order materials to jobsite, place bets on sports, “tell me a dirty joke” ect…you know…everything you do working on a jobsite.
My requirements:
• The open source ai model LLM doesn’t need to be cutting edge (Llama),
• Retrieval augmented generation (local)
• high recall in its memory (supabase),
• cost effective API fees (open ai?)
• option for speech to text (?)
• access (api) to multiple tools/calculations
• ability to run offline (Llama?)
• generative ai – original content (Pedantic framework?)
I’m not a great coder, so I prefer no code options. I’m a big believer in agile development so I want to be able to iterate on agents and improve them over time. I want to spend 80% of the time on the jobsite learning the job and finding areas where the agent can help and 20% of the time building the agent. Should I put in the upfront time building an ai agent builder or just build in N8N? Based on the list above, could you provide the best tools for me to use. I have watched a lot of your videos…I would be great if you could narrow it down for me.
Many thanks!! - Dave
Hey Dave, this sounds awesome! I’d definitely recommend getting started straight away building these out in n8n.
For your open source LLM: I’d recommend Qwen 2.5 Instruct or Llama as you mentioned
RAG - Qdrant or Supabase
Memory recall - Supabase
Cost effective API - OpenRouter
Speech to text - take a look at Whisper
Running offline - you’ll need to have a GPU powerful enough to run your models. I would do some research on hardware requirements for different LLM sizes, and test with those LLMs to see if they get the job done for you!
Pydantic AI is certainly not a no code option, so if you want that I would stick with n8n!
Great! This definitely helps narrow down my tools… which is half the battle. N8N is the platform I feel most comfortable with and it’s very similar to other no code applications I have worked with in the past. I’m hyper focused on getting product/market fit before I spend too much time in the development weeds. I started using elevenlabs last night, however I will check out Whisper for speech to text. I still need to find a good GPU that’s affordable and can run with low latency.
Again, many thx!
David