@habemus-papadum/aiui

ai ui frontends — the aiui CLI: launchers and browser plumbing for an agent-in-the-loop UI workflow. ⚠️ Read the repo's Read before running guide first: aiui claude skips permissions by default and gives the agent a browser.

Install

npm install @habemus-papadum/aiui

CLI

aiui demo      # scaffold a disposable demo playground (safe to re-run — it continues)
aiui claude    # launch Claude Code wired with the aiui channel, plugin, and browser MCP
aiui vite      # launch Vite against the running channel (sets VITE_AIUI_PORT)
aiui browser   # start (or find) the shared session browser; --tunnel <host> = remote-dev local half
aiui open <url># open a URL as a tab in the session browser
aiui chrome    # manage the browser: install | update | status | extension
aiui mcp       # forward to the aiui-claude-channel CLI (e.g. `aiui mcp quick`)
aiui --help

Flags for aiui claude beginning with --aiui- are consumed by aiui (--aiui-no-chrome, --aiui-chrome-profile <name>, --aiui-tag <tag>, …); everything else forwards to claude verbatim. --help/--version are inert: aiui prints its own answer, then the wrapped tool's follows — no config, browser, or Chrome-for-Testing activity. Durable settings live in config.json (user cache + project .aiui-cache/) — see the repo's Configuration guide.

Built with commander for the command tree and execa to spawn the child processes. The command implementations live in src/commands/.

During development, run the CLI straight from source (via tsx, no build) with the ./aiui launcher at the repo root:

./aiui claude
./aiui --help