Local AI Masterclass: LLMs, Diffusion & AI-Agents on Your PC

Explore the basics of local AI, hardware to run LLMs and diffusion models on PC and Apple devices, and how diffusion and LLM training with transformer architecture work.

Assess why local AI matters by comparing the disadvantages of closed AI (data privacy, leaks, costs) with open source benefits like offline operation, customization, and control.

Identify the hardware essentials for running local LMS and diffusion models, highlighting GPU VRAM (6–12 GB min, 24–48 GB ideal), RAM, storage, CPU, cooling, and power.

Apple devices enable efficient local AI with unified memory and on-device optimization, but lack CUDA and RAM/VRAM are soldered, limiting upgrades.

Learn how diffusion models generate images by converting prompts to embeddings and tensors, adding noise in forward diffusion, and denoising with rgb color codes through many steps.

Explore how LLMs such as GPT, Qwen, Deepseek, Llama, and Mistral are built from two files: a parameter file and a run file, using llama two as an example.

Learn to run AI locally, keeping data on your device for offline reliability and full control over fine-tuning, prompts, diffusion models, LLMs, hardware needs, and context windows.

Learn to run LMS on your device with Olama, install models, and test them with the Olama app. Master prompt engineering, vision with OCR, function calling, and quantization concepts.

Install Ollama and download the appropriate models, then verify your hardware VRAM before running the server and calling the endpoint from the terminal. Explore model options.

Explore the Ollama app for testing LLMs locally and master prompt engineering techniques, including role prompts, few-shot prompts, chain-of-thought, and interactive refinement.

Explore multi-modality in vision models, compare open-source and closed-source systems like Gemma and Qwen vL, and learn how OCR extracts text from images and invoices on your local PC.

Discover how large language models use function calling to access external tools, call APIs, generate images via diffusion models, search the web, and manage emails and databases.

Explore test-time compute and model thinking in LLMs, with step-by-step reasoning, chain-of-thought prompts, and fast vs slow thinking, highlighting math, coding, and science applications.

Delve into RAG and embeddings in Ollama, covering embeddings models for vector databases and options to run locally or in the cloud, plus basics of quantization.

Explore quantization of large language models and diffusion models, turning 32-bit weights into FP16, BF16, or 4–8 bit formats to save memory and boost speed on consumer hardware.

Install and use Olama to run and manage models locally, both in app and terminal. Explore prompts, context, file uploads, embeddings, function calling, and quantization options for consumer hardware.

Build rec applications with a nice Olama interface, covering vector databases, embeddings, models, top-k, chunking, and retrieval augmented generation, plus using anything LLM and LLM Studio.

Upload PDFs, convert with embeddings models into vectors, store in a vector database, and perform chunking with top-k retrieval for retrieval-augmented generation.

Optimize RAG by preparing data with markdown and Q&A, storing only relevant information. Convert messy data to markdown using HTML-to-markdown or Llama parse, then optimize retrieval with proper chunking.

install anything locally to build apps and ai agents with the llama model, and configure providers, Langs db, the anything LLM embedder, and 1000 token chunks with 20 token overlap.

Learn to chunk documents into embeddings using a text splitter, set chunk sizes and 1–5% overlaps for a vector database, and tailor ranges by text type.

Build a local rag chatbot with anything llm and ollama by embedding pdfs and other documents, optimizing chunk size and overlap, and retrieving with citations in a workspace.

Explore configuring system prompts, top-k results, and rack-app memory (not the vector database), tune temperature, and optimize similarity search and chat history settings.

Explore agent skills in anything lm, activate long-term memory, view and summarize workspace documents, conduct web search, create charts, and build local ai workflows with agent flows on your pc.

Explore LM Studio as a local all-in-one solution for running models with GPU offload, model search, and no endpoints, plus quick setup and context protocol basics.

Ornit 1.0 is an open source self-scaffolding llm for argentic coding, available in 9b, 31b, 35b moe, and 397b, with strong benchmarks and consumer hardware friendliness.

Learn how llm function calling retrieves information from a vector database using embeddings and a similarity search, and manage chunks, prompts, memory, and diffusion basics in the local ai masterclass.

Explore diffusion models and run them locally, install conf ui and manager, and master workflows to generate pictures with prompt engineering, control nets, Lora, Flux, and op mart.

Discover open-source tools to run diffusion models locally, including automatic1111, forge, focus, swarm UI, and conf UI. Learn why conf UI is the root for flexible, up-to-date ai workflows.

Install git and comfy ui on Windows or Mac, using one-click setup or GitHub manual install, ensure GPU compatibility, and understand convoy manager updates.

Install the conf ui manager to simplify config and workflow management in conf ui; locate the confi ui folder, run the installation, restart config, and manage models and custom nodes.

Learn ComfyUI basics: find and run your first workflow, download and load a diffusion model, and generate images using prompts, seeds, the interface, and positive/negative prompts with samplers and steps.

Explore basic workflows by browsing templates for image, video, audio, and 3D models, and learn to run locally or via API using diffusion, Laura, and a text encoder.

Explore running Stable Diffusion Excel locally with conf UI, using JSON workflows, prompts, and metadata to generate high-quality images on low-VRAM hardware.

Explore prompt engineering for diffusion models such as SDXL and flux, covering resolutions, sampling steps, cfg, weights with brackets, and how to craft effective positive and negative prompts.

Explore the open source queen image edit model to perform image editing, inpainting, outpainting, and depth, canny, and open-pose control nets, enabling local, gpu-efficient diffusion workflows.

Learn LoRA basics, a low-rank adaptation that trains a small part of a diffusion model and applies Laura on top to adjust style and detail.

Train Laura with Flux, using the FP8 base model for streamlined image editing workflows. Explore prompts, samplers, and setup to run Flux efficiently on your PC.

Train your own flux lora to customize logos, specific styles, animals, or faces, including an ai influencer, using local flux gem or cloud options like replicate with an h100.

Explore upscaling workflows in ComfyUI with Flux and SDXL, including one-step Flux upscaling, FP8 options, diffusion models, and control nets for crisper, more detailed images.

Learn how to find, download, and run comfyUI workflows on Opencart, rent GPUs, and use V3 models while navigating community content and safety notes.

Master local diffusion with comfyUI for running models on your PC and managing workflows and loaders. Use prompts, control nets, and LoRA to generate high-quality images locally.

Master local ai video generation using conf UI and frame pack with api access to OpenAI and Alibaba models. Explore prompts, videos from text or from images, and character swaps.

Learn to generate videos in ComfyUI with an API and prompts. Explore text-to-video and image-to-video workflows, Sora models, costs, resolutions, and prompt anatomy.

Explore local video generation with comfyui and wan models, leveraging open-source 12.2 and 2.2 animate variants, image-to-video workflows, and quantized options for hardware-limited setups.

FramePack enables 2-minute uncensored ai videos locally on a pc with six gigabytes of vram, using a 13b model and anti drift sampling for consistency.

Explore AI video generation using convoy workflows and various providers, from OpenAI to Google, with options for text- or image-based videos, open-source Gwen models, and AI agents.

Transcribe spoken words using Whisper from OpenAI, showcase a local text to speech solution OpenAI Audio, demonstrate AI music generation and voice cloning with closed source and open source options.

Transcribe locally with open-source whisper via the Pinocchio interface, using whisper large v2 for fast, private transcripts with timestamps and translation.

Explore local text-to-speech and voice cloning on your PC with open audio (fish speech), offering multilingual support, offline CUDA-enabled performance, and a Gradio interface to test emotions and reference audio.

Learn to generate AI music locally on your PC using Tsuno, Zugno AI, and other state-of-the-art models, via templates and a config UI for text-to-song, instrumental music, or editing.

Transcribe audio locally with Pinocchio and Whisper, then generate audio from text using a local tts model and explore voice cloning and AI music tools.

Explore ai agents and ai automations on your local machine using the nan low-code tool, docker, and supabase for sql and vector store, build an email assistant and refine prompts.

Install n8n globally on your machine using Node.js, npm install -g, or Docker, then explore the interface and template workflows for automations and AI agents.

mac users learn to download node and set up n8n locally via terminal using sudo npm install -g, update -g, and start, then open the app at the given url.

Learn to manage node versions with nvm for Windows to fix neidan installation issues, requiring admin on your local machine, and use nvm list, install, and use to switch versions.

Learn the core n8n basics—design triggers and actions using nodes, explore MCP servers and webhooks, and connect tools like Gmail, Google Drive, and AI agents to build modular workflows.

Install Docker Desktop to create isolated containers that prevent program interference and manage containers, images, and volumes. The next video installs Supabase on Docker to run it locally.

Install supabase locally with docker, clone the repo, and spin up a local instance with four commands; explore vector databases, embeddings, and a table editor for building an AI agent.

Build a local RAG workflow using an LLM agent with memory, Supabase vector store and Ollama, embedding data from PDFs, upserting to vector store, and querying with metadata filters.

Export and import JSON workflows to share and reuse across tools, demonstrating downloading a JSON file, opening it, and importing it to recreate settings, nodes, and configurations.

SQL tables with Supabase and n8n: create and query data, upsert rows, import CSVs, list or delete records, and enable AI agents to access and manipulate the database.

Build an email automation agent that queries a database, retrieves contacts, and sends emails via Gmail using n8n and Supabase SQL integration.

Master system prompts for ai agents by defining role, goal, tools, rules, style, and output format, then optimize and test prompts using examples and a prompt optimizer.

Explore how n8n enables local automations with email triggers, ai classification, and drafts using a local model like Olama, plus templates, vector databases, and Supabase integrations.

Install NodeJS, set up Argentic II with triggers and AI agents, and build local vector and SQL databases with Docker and Supabase, enabling email automation.

Master MCP workflows by setting up an MCP server and a client, connecting Elm Studio and Open web UI, and triggering databases, email automations, APIs, and OCR tasks.

Build a local MCP server and client with the open source model context protocol to enable function calling, connecting AI apps to external tools and databases using LM Studio offline.

Learn how to use LM Studio as a frontend for an MCP server to drive n8n AI agents and connect tools like Supabase, a vector store, and CoinGecko.

Learn to automate local image and video generation by connecting comfy UI with n8n, building workflows, and using chat triggers, AI agents, and Google Drive uploads.

Learn to use LM Studio as a front end for Conf UI, create subworkflows in n8n triggered by another workflow, and optimize testing for AI-driven image generation.

Install and run Open Web UI locally with Docker, configure Olama, and explore the interface, settings, models, webhooks, and pipelines.

Explore open WebUI basics with whisper dictation, voice chat, a code interpreter, and web search enabled by an api key, plus embedded PDFs.

Learn to use Open Web UI as a frontend for an MCP server, configure external tools and bearer authentication, and connect via webhook-driven pipelines to run AI agent workflows.

Learn to automate image generation with HTTP requests in Open Web UI using n8n and Replicate, map prompts, authenticate with API keys, and save results to Google Drive.

Explore OCR with vision models Gemma 312 B and Quan 2.5 VL extract data from images, invoices, and PDFs, export to CSV or save to Google Drive or Google Sheets.

Explore a complete local social media content machine on your PC that schedules hourly workflows, generates video prompts, adds audio with FFmpeg, and uploads videos to YouTube.

Assess the value of fine tuning with lora, noting risks of overfitting, cost, and slower updates, and why a strong system prompt can suffice.

Leverage the MCP protocol to translate API calls between server and client. Connect servers to any host, share databases across realms, and build automations with open web UI.

compare knn and flowwise to understand advantages, install flowwise locally with node.js, set up postgres in docker, upsert vector data, and create a local rag agent with workflows.

Explore how LangGraph, LangChain, and Flowise differ and integrate to build AI agents, with LangGraph creating agents and LangChain powering chat flows in the background.

Install flow wise locally with Node.js by running npm install dash flow wise globally, then mpx flow by start to launch on localhost:3000 and explore its open-source AI agent interface.

Learn to install a local Postgres vector database with Docker, run a bg vector db container, configure postgres, and expose port 5432 for local applications.

Set up and populate a Postgres vector database with uploaded PDFs, configure embeddings and document loaders, and manage chunks with a record manager for deduplication and efficient retrieval.

Build a local rag agent with Flowise, Postgres vector store, and Ollama to answer from documents using memory and tools; embed the chatbot on a web page.

Explore Flowise marketplace and community templates to unlock more AI agent workflows. Learn about agent flows, vector databases, human-in-the-loop checks, and addons for curl-based requests.

Install flow wise, use LangChain concepts, set up Postgres in Docker as a vector database, and build a rack application with agents version two using llms, data loaders, and tools.