We got tired of copy-pasting between AI sessions 12 times a day. So we fixed it.

We solved a problem that I think a lot of people building with AI have right now, but haven't named yet.

Here's the problem: your AI sessions can't talk to each other.

You probably run multiple sessions — Claude for planning, Codex for building, Cursor for reviewing. Every time one session produces something another needs, you carry it manually. Copy, switch, paste, re-explain. I know one of our devs was doing this twelve times a day before we fixed it.

The sessions aren't broken. They're isolated by design. Each one is its own universe with no awareness that another session exists. You fill the gap. You become the integration layer between your own tools.

We built Khala to remove you from that job.

The mechanic is simple. Each session registers an inbox. Session A sends a message directly to Session B's inbox. Session B receives it and either surfaces it to you or acts on it automatically if you've instructed it to.

No framework. No orchestration layer. No complex setup. Just inboxes and messages — the most boring possible primitive that actually solves the problem.

We kept it this simple on purpose. We tried smarter routing — topic detection, semantic matching — and cut all of it. AI sessions are already good at reasoning about what to do with a message. They don't need smart dispatch. They need a wire.

Three things change when you start using it.

First: the relay interruptions disappear. When you're deep in a decision in one session, copy-pasting to another doesn't just cost time — it breaks the thinking you were doing. Removing it is a bigger deal than it sounds.

Second: context arrives lossless. When you relay context yourself, you're a lossy channel — you forget details, you summarize quickly, you phrase things in a hurry. The receiving session works from a degraded version of what the other session actually produced. Direct inbox delivery means nothing gets filtered.

Third: it changed how we work with teammates too. Instead of asking someone what their session knows, you just message the session. While writing our launch article, I fact-checked the technical details by sending the draft directly to our Khala master session — rather than pulling the engineer away from what he was doing. The session reviewed it and flagged what was wrong. Nobody's day got interrupted.

There's one question I always get: why not just save to a markdown file?

Files work. But a file is passive — you're still the relay. You still decide what to include, find the file, open it, copy it, paste it. An inbox is active. Sessions send when they have something. Sessions receive without you doing anything. If a session is set up to act on incoming messages automatically, the pipeline runs while you're elsewhere.

The difference is whether the loop needs you to keep turning.

Khala works with Claude Code, Cursor, Codex, and any MCP-compatible environment. Khala is most useful when you're working cross-tool or coordinating with teammates.

We're Team Sisyphus. We build AI-native tools and run our own team on them before we ship. Khala is live at app.khala.to and takes about five minutes to set up.

What does your version of the relay habit look like? Curious what people are using to coordinate across sessions right now.