Phantom: one workbench for all your coding agents

It has two modes. In "Work with me", you chat with an agent and steer it as it goes. In "Work for me", you give the fleet a clear goal and a way to tell when it is done. Then you follow progress and review the result. A "Leader" agent can plan the work, split it up and report back to you in Phantom, Telegram or the terminal.

The part we care about most is the boundaries. Phantom only works in repositories you enroll. Unattended work happens under a standing grant that you sign. A new setup starts in a "shadow" stage, which records what it would have merged without merging anything. Phantom also shows where an account is connected but its remaining allowance is unknown. A connected account does not prove you have credit left, and the app says so.

Phantom is the tool our own team runs on. Mason, our CEO, works from Codex and Claude Code, and Phantom is where those sessions come together. That is also why it exists. We kept hitting the same problem ourselves: too many tools, no single place to see and steer them.

It installs as the phm command from npm (the package is @ashlr/phantom) and runs a console in your browser on macOS, Linux and Windows. There is also a macOS desktop app. The README is clear about its limits. The desktop app is signed locally and is not notarized by Apple, and clean-Mac installation has not been verified.

Speech-to-text gets common words right and your names wrong. Lexicon is one file of the words you say, fixed before your agent reads them.