Uploaded June 2026 | Updated November 2026, 13 hours ago
A livestream for everyone starting out with open models. We walk through the most robust stack for running AI on your own laptop or server, and answer your questions live.
What we cover:
Merve Noyan on llama.cpp: what it is and how it runs on your hardware, plus the new llama.app and llama barn.
Daniel Han from Unsloth on downloading GGUFs: finding quantized models on the Hub, why quantization matters, and choosing the right size and type for your hardware.
Ben Burtenshaw on selecting your harness: closed harnesses like Claude and Codex, open harnesses like Pi, with a demo using llama server.
Onur Solmaz on Pi PR triage: a demo of Pi and Gemma for automated PR triage.
Plus a live AMA throughout the stream.
Bring your questions about local models, llama.cpp, open coding agents, and anything open or local AI.
00:00 - Introduction to the livestream series
05:04 - Q&A: Are open models actually good? (Benchmarks vs. Public Opinion)
09:01 - Defining Local Models: Privacy, security, and cost benefits
10:44 - Llama CPP Deep Dive with Merve
13:30 - New Llama.cpp features: Unified CLI and Web UI overview
15:30 - Finding compatible models on Hugging Face Hub
18:30 - Demo: Generating a website with Gemma and Llama.cpp
19:35 - Introducing Llama MacOS: Managing VRAM and context windows
21:10 - Q&A: Unpacking hardware costs and model selection strategy
25:57 - Dynamic Quantization with Daniel Han: Making models smaller without losing accuracy
31:55 - Improving inference speed with Multi-Token Prediction (MTP)
33:00 - Exploring Gemma and its variants
35:55 - Live Demo: Creating a Flappy Bird game using local models
38:40 - Connecting coding agents (Claude Code, Pi, Open Code) to local models
41:45 - Demo: Comparing different agent harnesses on repo analysis
50:50 - OpenClaw and LM Studio Demo with Onur: Optimizing for Mac Silicon
55:50 - Performance Benchmarks: Tokens per second on local hardware
59:15 - Privacy vs. Security tradeoffs in local AI
01:03:50 - Final wrap-up
A livestream for everyone starting out with open models. We walk through the most robust stack for running AI on your own laptop or server, and answer your questions live.
What we cover:
Merve Noyan on llama.cpp: what it is and how it runs on your hardware, plus the new llama.app and llama barn.
Daniel Han from Unsloth on downloading GGUFs: finding quantized models on the Hub, why quantization matters, and choosing the right size and type for your hardware.
Ben Burtenshaw on selecting your harness: closed harnesses like Claude and Codex, open harnesses like Pi, with a demo using llama server.
Onur Solmaz on Pi PR triage: a demo of Pi and Gemma for automated PR triage.
Plus a live AMA throughout the stream.
Bring your questions about local models, llama.cpp, open coding agents, and anything open or local AI.
00:00 - Introduction to the livestream series
05:04 - Q&A: Are open models actually good? (Benchmarks vs. Public Opinion)
09:01 - Defining Local Models: Privacy, security, and cost benefits
10:44 - Llama CPP Deep Dive with Merve
13:30 - New Llama.cpp features: Unified CLI and Web UI overview
15:30 - Finding compatible models on Hugging Face Hub
18:30 - Demo: Generating a website with Gemma and Llama.cpp
19:35 - Introducing Llama MacOS: Managing VRAM and context windows
21:10 - Q&A: Unpacking hardware costs and model selection strategy
25:57 - Dynamic Quantization with Daniel Han: Making models smaller without losing accuracy
31:55 - Improving inference speed with Multi-Token Prediction (MTP)
33:00 - Exploring Gemma and its variants
35:55 - Live Demo: Creating a Flappy Bird game using local models
38:40 - Connecting coding agents (Claude Code, Pi, Open Code) to local models
41:45 - Demo: Comparing different agent harnesses on repo analysis
50:50 - OpenClaw and LM Studio Demo with Onur: Optimizing for Mac Silicon
55:50 - Performance Benchmarks: Tokens per second on local hardware
59:15 - Privacy vs. Security tradeoffs in local AI
01:03:50 - Final wrap-up









