Tom
A local Windows AI with its work in view
Start new thread

Tom - A local Windows AI with its work in view

by•
AI for more of us. Tom brings local Gemma 4 to everyday Windows PCs, with a light harness, visible actions and local file tools. No dedicated GPU required. Built with GPT-6 Astra. Early beta: browsing and complex tasks still need patience.

Add a comment

Replies

Best
Maker
📌

Hi Product Hunt! I’m Dekker, building Tom at OsbornVentures. I believe in AI for all. To get there, we have to start with the computers people already own and make the experience easier to understand.

Tom’s design target is ordinary Windows PCs from the DDR4 era onward, without making dedicated graphics memory the price of entry. Conservative settings use a small CPU thread budget. The aim is useful local AI on everyday hardware, with Windows integration that feels at home. Physical 8 GB testing is still pending, so this is a design target rather than a promise for every machine.

Gemma 4 brings language, coding and image understanding; Tom adds a lightweight harness to put those abilities to work with local files, programs and visible actions. Nine animated ASCII states, an informative startup checklist, a context ring, expandable activity and offline AI help explain what is happening. Setup asks what Tom should call you, and file results open locally.

Please be patient with Tom! The harness is a work in progress and draws on elements of our recent Eldon harness. Tom can be very useful for focused local tasks, but browsing is not a strong point. Complex code and long tasks still need supervision. Our small release checks passed 5 of 6 focused local tasks and 2 of 5 broader attempted tasks on one 32 GB Windows PC. The public report includes the failures; these are not general reliability scores.

GPT-6 Astra helped me build, test and refine Tom’s app and harness. This is our GPT-6 Astra Challenge entry; Gemma 4 runs locally through llama.cpp when you use Tom. Thank you to the teams behind the stack—credits and licenses are included.

Tom is free, with Apache 2.0 source, one complete full installer and a small network installer. This release is unsigned. The website lists prerequisites and checksums to help you get started.

What everyday task would you want Tom to make easier?

OpenAI
GPT-6 Astra Challenge
How did Astra change the scope or ambition of what you built?
Maker
Astra helped me pursue a complete Windows experience around local AI: a lightweight harness, visible actions, animated ASCII states, startup progress, context education, local file links and installers. It assisted with implementation, testing, documentation and release work, including a single hosted offline EXE and a small network installer. That supports Tom’s larger aim: AI for people using everyday Windows PCs, without requiring dedicated GPU memory. The harness draws on our Eldon work and remains a work in progress. Astra also helped make the limitations reviewable: we published small benchmark results with failures rather than claiming universal reliability. Gemma 4 powers Tom locally; Astra helped build and refine the experience around it.