What is Janissary?

Janissary is an Agentic Working Environment: an app for directing AI agents rather than typing every command yourself. The core interaction is the agent — you open one, give it a task, and it works in its own space while you move on to something else. Each agent communicates through a tab: a text-based transcript where you send it commands or prompts and read what it does in return.

The janus command starts it. A session begins as a single tab and grows as you need it to: agent opens a fresh agent tab with its own transcript, shell, and working directory; harness claude puts a full AI coding harness in a tab; open, edit, and files open viewers for images, Markdown, code, and directory trees alongside your agents. You type at one command bar; what you type can be an application command, a shell command, or a prompt to an agent, and the app works out which you meant. Tabs stay independent — a long build in one tab doesn't block a conversation with an agent in another — and the tab strip shows at a glance which agents are busy and which have new output waiting.

The tab is the core idea. Everything else builds on it: groups keep related tabs together, scheduling runs commands in a tab later, agents message each other across tabs, and profiles recreate a whole multi-tab setup with one command. For work you'd rather keep at arm's length, a workspaced agent gets a disposable clone of your repository, isolated from the rest of your machine.

If you're new, start with Starting the app, then read Tabs — the rest of the docs assume both.