My day job is software engineering, where I spend a lot of time in tools that are fine but never great. The little frustrations add up, and eventually my curiosity got out of hand. Meet Threa.
Three of them stuck around long enough that I did something about them.
Years of decisions and their reasoning pile up in workplace chat, and I rarely find them again. My memory runs on gist: I'll remember that we discussed something, sometimes why, rarely the exact wording or where we said it, though I'll know it was Kate who said it. I started wondering whether a chat app could do something more useful with all that conversational data, keeping decisions and their reasoning tied to the messages they came from. I assumed it would cost too much. So far it hasn't.
Most tools want to become the platform you can't afford to leave, and lately plenty of them insist on shipping their own AI agent. I rarely want it. The agent I already use has my codebase, my tools, and the context I've built around it; another agent inside another app starts without them. Threa has Ariadne too, and I want her to be genuinely great, but she's at her best inside the room, where the shared context lives. For my own work I'd rather bring the agent I already use, so Threa connects to it through the bot runtime, public API, CLI, and MCP server instead of trying to replace it.
Nested threading, quoting, replying to one message in a busy room, formatting that behaves. None of it sells a platform, which is probably why it keeps getting skipped, but it all adds up over a day spent in chat. After years of putting up with the basics done badly, I think we deserve nice things.
Hi, I'm Kristoffer. People have paid me to build software for over ten years now. During parental leave I started tinkering on Threa to keep myself from going insane, and I'm not sure it worked: here I am, up late after my wife has gone to bed, talking to an agent about who I am and why I'm building this.
Good at its job, composable with everything else, and easy to step away from. Threa is built that way on purpose. The door stays open, so if you stay it's because you want to, not because leaving would cost you.
It has also been my way into building with AI agents. Most of Threa is written by directing them, but the thinking stays mine, and that matters more than ever. They're a good way to chase a curiosity, and you can get genuinely decent software out of them as long as you do the thinking yourself. Seeing where they shine and where they fall short has been half the fun. It's a work in progress, on what I think is a good trajectory. The repo is public, so have a look.
Threa is early, and there's plenty left to build. Join the waitlist, or follow along in the public repo.