About Shannon
An always-on agent that works in your documents.
Shannon is an AI productivity agent. You hand it a task; it edits your documents directly in the cloud, streaming every step as it goes — and every change comes back as a tracked edit you can accept, reject, or rewrite. And because it runs in the cloud, it keeps working after you close the tab. This site is the engineering journal of building it.
How it’s different
- Files form the foundation. The agent’s response is the document itself. It doesn’t describe an edit in a chat and hope you apply it; it makes the edit, and the file in front of you is the deliverable.
- Always-on agents. Shannon runs on a server around the clock, so work continues after your laptop lid closes, and a task can come from anywhere: the app, or a mention on X that pulls a summary into your workspace.
- Context belongs in the cloud. Every channel and every model reads and writes the same workspace, so nothing you save dies in a chat thread. The longer you work in it, the more the agent has to work with.
- Transparency builds trust. Its thinking and tool calls stream live while it works. Every edit lands as a tracked change you can accept, reject, rewrite, or branch, and version history puts any file back the way it was. If you can’t see what an agent did, you can’t trust it with your documents.
- Mixture of models. No loyalty to a single model: each task goes to whichever one does it best today, and when something better ships, the swap is routine.
What you’ll find on this site
- Engineering on the hard parts of an agent that edits documents: the agent loop, model routing, streaming, and the review pipeline.
- Lessons learned written the way I write them internally: what happened, what held, what I changed.
- Product updates in plain language, so you can follow what shipped without knowing the stack.
Who’s building it
Tuan Huynh