What we tried to make CI faster
What worked in Char's CI, why we reverted extra workers, and the cost of getting feedback sooner.
Notes from the Char team
On product, local-first software, AI workflows, and the engineering behind an AI notepad that knows and works for you.
Making design decisions survive the next session
Part 2 of vibe coding a consistent product. How we connected design.md, Storybook, and a reuse gate so coding agents could find and follow our design decisions.
Abstraction is beautiful
Unifying Cards, Chips, and Prompts in Char, then asking whether a custom Schedule can reproduce what our built-in Cards do.
Our updates were slow because we asked 617 times
Char's desktop updates took minutes to download bytes that need seconds. The problem was never bandwidth. It was one request at a time. Here's how we went from 286 seconds to 12.
Keep your agent's brain outside the box
Part two of building Charlie's cloud, and the decision part one hinged on. The agent loop and the keys stay on our servers, the code runs in a sandbox, and two machines taught us that routing is all or nothing.
Pick your agent's computer by design, not by price
What we learned giving Charlie its own computer. We tested Boat, Fly and E2B, priced thirteen sandbox vendors, and found that where your agent loop lives decides the box.
Tom Riddle's diary
The best way to describe Char right now — you write into it, and it writes back. Here's what that means, and where meetings fit now.
One page that remembers
We are replacing Char's stack of daily notes with one continuous scratchpad that carries unfinished work forward.
Who manages the managers?
The Slack-DM-to-self hack, why founders are becoming managers of agents, and why Char starts as a daily note instead of another meeting bot.
After the meeting ends
A year ago we launched Char as an AI meeting notetaker. We came out the other side believing the real problem was never how you capture a meeting — it's what happens after. Char is built around that answer.
What Char is building
Char is the AI notepad that gets things done. Open loops carry forward on Home, meeting context stays close, and delegated work returns to the page that started it. It looks deliberately simple. Underneath, Char is building the memory and handoff layer that lets people direct agents without turning every thought into a new chat. Everything stays close to the work instead of becoming another inbox to manage.
This blog is where we explain the product as we build it: what an AI notepad should remember, where it should forget, and how a daily note can grow into something closer to a chief of staff. We write from the work itself, including the decisions that held up, the ones we reversed, and the engineering required to make persistent work memory useful without making it invasive.