The problem Freedom AI solves

Most AI browser tools rely on centralized cloud APIs, which means prompts, browsing context, and task data leave the user’s machine. Freedom runs a local AI agent in the browser and can coordinate work with other browser agents through Swarm. Users can ask it to research pages, summarize content, or perform browser tasks while keeping more control over where data is processed. Larger tasks can be sent to other agents, with results written back to Swarm as content-addressed data. Agents can also be selected by capability and, later, by cleaner available electricity for inference and other sustainability markers.

Challenges we ran into

We uploaded a Gemma4 AI model family to Swarm, to allow for P2P downloads to Freedom users around the world. The main model file is 7 GB, which is a real challenge to upload and persist on Swarm. We extended our own Swarm node implementation to allow for stable and fast uploads of large files.

To let Freedom instances speak to each other for sharing work other across networks, we used the XMTP protocol. Messaging via pub/sub for light nodes of decentralized storage networks is not a solved problem yet.