Build a Capable AI Assistant Without the OpenAI API — Entirely Local

There are three main reasons developers choose to build without the OpenAI API: cost (no per-token charges), privacy (no data sent to OpenAI), and independence (no risk of API changes or price increases affecting your assistant). This workshop delivers all three by using free open weight models running locally through Docker Model Runner.

Docker Model Runner replaces the OpenAI API in this workshop. It runs free open weight LLMs locally on your machine and exposes an OpenAI-compatible API endpoint. OpenClaw connects to this local endpoint exactly as it would connect to the OpenAI API — but with zero cost and complete data privacy.

Yes. Modern open weight models available through Docker Model Runner handle a wide range of complex tasks including code writing, analysis, summarisation, and extended conversation. The instructor covers model selection to help you choose the best model for your specific task requirements during the live workshop.

The most notable difference is that local open weight models generally have knowledge cutoffs and may not have the very latest information. They also do not include image generation by default. For conversational AI assistant tasks, the functionality gap between local models and the OpenAI API has narrowed significantly in 2026.

Updating your local AI assistant is straightforward — you pull a newer version of your chosen model through Docker Model Runner when you want to upgrade. This is simpler than managing OpenAI API version changes and entirely under your control. The instructor covers the model update process during the workshop.

Yes. The open weight models used in this workshop have licences that generally permit commercial use. The instructor covers the relevant licence considerations during the live session. OpenClaw itself is open source with a permissive licence.