Cross platform companion for multi agent

Burmese is a cross platform companion for multi agent work, published at theburmese.xyz and discussed via HN AI Agents. Its author built it after running many parallel coding agents and finding that answering “is it done yet?” meant hunting across terminals and tmux panes. Burmese is meant to replace that scattered check with a single place to see what is running.

Technically, Burmese natively discovers agent sessions already running on the machine. It targets Claude Code, Codex, and other harnesses. It does not copy or fork those sessions. It resumes the real conversation by the harness’s native session ID, so the control surface attaches to the same session the agent already owns rather than a duplicate.

For engineers and builders who keep several coding agents in flight, the useful part is operational, not novelty. Status and progress stop living only inside individual panes. Sessions, projects, and agents are meant to land in one live control center, so a glance replaces pane-by-pane hunting when you need to know what is idle, blocked, or finished.

In market terms this sits next to the rise of agent harnesses such as Claude Code and Codex, where the bottleneck is less “can I start an agent” and more “can I supervise many at once.” Most workflows still treat terminals and tmux as the UI. Burmese assumes those harnesses stay the system of record and adds a discovery and resume layer on top instead of replacing them.

Practical takeaway: if you already run parallel agents across terminals, the product to evaluate is whether native session discovery and resume by session ID actually surface every active run you care about in one control center. Watch next for how cleanly it keeps project and agent grouping accurate as you add more harnesses beyond Claude Code and Codex, and whether resume stays reliable when sessions are long-lived or restarted outside the UI.