Open Weight Agents

You picked up this book because the headlines will not stop. Kimi ships a trillion parameter model, DeepSeek publishes its reasoning weights, GLM drops an MIT licensed giant, Qwen powers 1000s of deployments, and Llama makes open weights a household name. Someone says open models now rival Claude, someone else says you still need the expensive tools, and you are left paying 2 subscriptions while wondering if any of this is actually usable for the work you do.
Or maybe you want to cut costs, protect privacy, or run a model on your own machine without asking permission from a vendor. You do not need a PhD and you do not need another benchmark table. You need a practical map that translates hype into a stack you can run today on hardware you already own.
This book is not a worship of open weights and it is not a dismissal of Claude Code, Codex, and Cursor. It is not a promise that 1 model wins at everything, and it is not a tutorial that assumes you already code for a living. What you need is a clear, honest framework for choosing, running, and verifying, with the cost and privacy questions answered up front.
Inside, you will discover:
• What open weight really means, and why it is not the same as open source
• The 5 families that matter, and the job each 1 is actually built for
• How to separate the harness from the model and run open models inside closed tools
• A 4 step build loop that turns any model into results you can verify
• A 5 axis decision framework for picking without paralysis when releases fly weekly
• How to run quantized models on a laptop with 16 GB to 64 GB of memory
• The true cost math, from $0 local to cheap APIs to premium subscriptions, so you stop overpaying
• How to build a small research agent that reads and organizes your documents privately, offline