Making Software
← All Episodes
Episode 10·October 28, 2026·00:52:41

Episode 10: Running AI on Your Own Terms: Local Models and Open Source

Show Notes

What if a model running on your own laptop could do most of your agentic coding, at close to the speed and quality of a top cloud model?Deepu K Sasidharan is a Principal Developer Advocate at Auth0 and has been building and maintaining open source for around fifteen years, including as co-chair of JHipster. These days he writes most of his code in Rust, and he built LlamaStash, an open source tool for running local models. He went from skeptical about AI coding to doing about 80% of his work on a local model, and he shares the setup behind it.I haven't run a local model myself, so in this conversation I ask the beginner questions: what llama.cpp actually is, how Ollama and LM Studio fit in, and where someone like me would start. We also get into the fear of losing touch with coding, why Deepu says we're all engineering managers and architects now, how open source maintenance is changing, and the layers of security and privacy he puts around AI agents on his machine. What You'll Learn :- What llama.cpp is, how local models run, and how Ollama and LM Studio relate to it- How to get started with local models, whether you prefer the terminal or a GUI- How a local model compares with a top cloud model on speed, quality, and cost- How to set up a coding harness where a cloud model plans and reviews, and local agents build- How to secure AI-assisted coding in layers: rules, deny lists, hooks, a classifier, and keeping secrets out of the environment- Why using AI to code can leave you feeling more technical, not less, and what the engineer's role becomesResources:- Deepu's blog: https://deepu.tech/- Deepu on X: https://x.com/deepu105- Introducing LlamaStash: https://deepu.tech/introducing-llamastash/- How fast is LlamaStash? A comparison with Ollama and LM Studio: https://deepu.tech/benchmarking-llamastash/- Can Qwen 3.8 running on your laptop really replace a frontier model for agentic coding?: https://deepu.tech/local-ai-qwen3.8-pi-llamastash/- Qwen3.8-Flash-Next running locally for agentic coding: https://deepu.tech/local-ai-qwen3.8-flash-next-best-local-llm/- My fully offline AI-assisted Linux development machine: https://deepu.tech/my-fully-offline-ai-assisted-linux-development-machine/

Guest

Deepu K Sasidharan

Deepu K Sasidharan

Principal Developer Advocate, Auth0

Deepu is a Principal Developer Advocate at Auth0 (Okta), covering identity for AI agents. He has 16+ years in the industry, and before Auth0 he worked at Adyen, XebiaLabs, and TCS. He's the creator of LlamaStash, KDash, JWT UI, and JDL Studio, most of them written in Rust and running in a terminal. He's co-chair and vice-president of JHipster, a core maintainer since 2015, a Java Champion, and the author of Full Stack Development with JHipster (Packt, two editions). He runs local models for his daily coding setup and writes up the benchmarks and tuning that made it usable. He has given 80+ conference talks and is a long-time Linux user, currently on Arch with Niri.

Read the blog post →

Subscribe to Making Software