2026-09-10 –, Track Room 1
LibreThinker is a free AI assistant extension for LibreOffice Writer, with over 19,000 downloads in 9 months. It started as a Python project wrapping GPT's API and grew into a multi-backend tool that supports free hosted models, connecting your own AI provider, and running models locally on your own machine. This talk covers how it was built, the engineering problems that came with running a free public endpoint, and what users actually ask for.
When I set out to add AI features to LibreOffice Writer, I could not find an extension that worked out of the box. So I wanted to release a small Python extension that wrapped GPT's API. Building the sidebar was its own challenge; LibreOffice's sidebar APIs are sparsely documented, and I got unstuck by finding UNODIT, an open-source tool that generates the scaffolding code for a working sidebar from an exported dialog file.
Discovering OpenRouter opened the door to hosting a completely free service for users. That introduced async job handling, error recovery across very different model backends, and the operational realities of running a free public endpoint. Then came support for users who wanted to connect their own AI provider account (via LiteLLM), and the option to run models entirely on the user's own machine.
Nine months on, the extension has over 19,000 downloads. Alongside the technical story, I will share how the community received the project, the most common feature requests and issues, and the user experience decisions behind it.
Mihail is a software engineer based in the UK, specialising in Python and agentic AI. He is the developer of LibreThinker, a free AI assistant extension for LibreOffice Writer with over 18,000 downloads. He also created ForecastSheets, an extension for Google Sheets that generates time-series forecasts and has been used by more than 3,000 people.