Matthew Berman starts with Unsloth, which combines local model inference, fine-tuning and training with an agent-style interface, tool access, memory and remote control. He presents its point-and-click training workflowNo-code model training uses a visual interface and guided settings to prepare data, configure training, run jobs, and evaluate AI models without requiring users to write training code. as a lower-friction entry point for people who find model configuration intimidating.
Matthew Berman also covers Diagram Design and Obsidian Skills. Diagram Design gives coding agents reusable instructions for producing cleaner technical diagramsAn AI agent skill is a reusable package of instructions, resources, and tool guidance for performing a bounded kind of work., while Obsidian Skills helps agents work with local-first Markdown notes as a personal wiki or durable knowledge baseA personal knowledge base is an organized collection of notes, sources, links, and ideas maintained for one person's long-term retrieval and use..
Matthew Berman rounds out the list with Buzz, Ego Lite and Modly. Buzz treats people and agents as participants in one self-hostable collaboration system, Ego Lite gives agents a fast isolated browser, and Modly converts source images into local 3D meshesImage-to-3D generation creates a three-dimensional model from one or more source images using visual inference and learned shape priors. for games or printing.
Watch on YouTube
