FullDots — Open-Source Contributions

About the project

FullDots is an open-source, MIT-licensed workspace for running AI agents on your own machine. Each agent, called a Dot, has a role, instructions, a model, the tools it is allowed to use, and optionally its own computer: a Docker container with a browser, files and a shell.

It is local-first: conversations live in a single SQLite file, with no account and no telemetry. Risky actions such as sending, paying or deleting pause for the owner's approval, and tools connect through MCP.

My contributions

Guided Docker setup for agent computers

Agent computers run in Docker, which can be hard to set up across operating systems. I built a guided setup:

• a setup engine that runs fixed, per-platform commands without a shell, and only after the owner approves them;
• a setup panel in the interface, with approval-gated agent tools;
• verification on Windows 11.

Pull request: #38.

File storage for Spaces

I added file storage to Spaces so agents and people can work with files side by side:

• a viewer for CSV, Markdown, JSON, images, PDF and sandboxed HTML;
• agent tools to save and read files;
• chat cards that open a result beside the conversation;
• hardening with server-side file-type detection and a strict Content Security Policy.

Pull request: #42, touching about 50 files.

Why it matters

Giving an agent a computer and files is powerful and risky. These changes follow the project's approach: the owner stays in control, anything that runs is explicit and approved first, and untrusted content, such as an HTML file an agent produced, is sandboxed.

Links

• Repository: github.com/asasemahmed/FullDots