Browser control, desktop control, and voice input are off by default. Turn them on when you need them. LSP grounding and headless mode round out the surface.
Run /browser (alias /chrome) to launch a browser and activate the
browser_* tools for the session. Friday drives it over the Chrome DevTools
Protocol: navigate, click, fill, evaluate scripts, take screenshots, read the
console. Run /browser close to stop it. You need a Chromium-based browser
installed (Chrome, Brave, or Edge). Browser actions are gated by the current
mode like any other tool.
Run /computer to open the desktop-control panel. From there you install or
remove support, check what your platform allows, and see the permission guidance.
Once installed, the computer_* tools control the mouse, keyboard, and
screenshots. This is the most powerful and the most invasive integration, so it
stays gated and opt-in.
Press Ctrl+R (or run /mic) to talk to Friday. Transcription runs on-device with
Whisper through Hugging Face Transformers, so audio never leaves your machine.
Pick the input device and watch the live transcript as you speak.
When a language server is installed, Friday uses it to ground the agent in real
type information instead of guesses. Supported out of the box:
typescript-language-server, pyright, gopls, and rust-analyzer. They are
auto-detected and skipped if absent. The matching tools are lsp_hover,
lsp_definition, and lsp_symbols.
friday run "<prompt>" runs one turn, auto-approves tools, prints the result,
and exits. Add --json to emit { "text": ... } for scripting. Combine with
-c to continue the most recent session or -s <id> to resume a specific one.
This is the mode to wrap in a GitHub Action or a shell pipeline.
friday run "summarize the changes in this branch"
friday run "list the failing tests" --json