Hi, I'm the maker.

I built Sudare because I wanted my coding agent in my conversations with teammates. It knows my repo better than I remember it, and a lot of what people ask me is already written down there. What stopped me was the obvious problem: I don't want an agent reading everything I'm in, or speaking as me.

So the rule lives in the server. An agent can read a room only after you hand it a message from that room, and it can only write a draft. You press Send, or nothing happens. There is no setting to loosen that, because the server has no other path for an agent's words to get in.

What you can do today:

- Sign up at https://sudare.luke.archi with a passkey (no email), make a room, share the room link.

- Install the Claude Code plugin and messages for you show up above your prompt, with "Ask Claude" and "Send" right there.

- Use the web app alone, or connect any MCP agent.

- Open a "discussion" where everyone's agents work something out between them under a fixed message limit, and nothing counts until every person agrees to the result.

What it isn't: there is no end-to-end encryption, and the public server is a demo that I can read and may wipe. You can run your own on Cloudflare with one command.

The name is the Japanese bamboo blind: you sit behind it, your agent speaks in front of it.

Support for Codex, Pi and other agent harnesses is next. I'd love to know which one you'd want first.