saynow

Speak text aloud from the terminal. Built so LLM agents can talk to you.

saynow "the build finished, 42 tests passed"

Works with no configuration — it falls back to your OS's built-in voice, which is offline, free, and needs no API key. Add a key for better audio.

Install

pip install saynow          # or: uv tool install saynow

No third-party dependencies — standard library only. An identical npm package exists for Node; both read the same config file, so pick whichever runtime you already have.

Usage

saynow "text to speak"              # speak an argument
echo "text" | saynow                # speak stdin
pytest 2>&1 | tail -1 | saynow      # speak a command's last line
saynow -p openai -v nova "hello"    # pick a provider and voice
saynow --save note.mp3 "hello"      # write a file instead of playing

If a cloud provider is configured but its key is missing, saynow warns on stderr and speaks with the system voice anyway. Use --strict to opt out. Concurrent calls are serialized so speech never overlaps.

Configure

saynow init                          # interactive setup
saynow config set provider openai
saynow voices                        # available voices
saynow models                        # OpenRouter models that can speak

Config lives at ~/.config/saynow/config.json, mode 0600. Precedence is defaults → config file → environment → flags. API keys are never accepted as flags, since argv is visible to ps and lands in shell history.

Use it from an agent

Add this to your CLAUDE.md, AGENTS.md, or equivalent:

You can speak to the user out loud with `saynow "<text>"`. Use it for things
worth interrupting for: a long task finishing, a question that blocks
progress, or an error that needs attention. Keep it to one short sentence —
it is spoken aloud, not read. Do not narrate routine progress.

Asking, and showing

--ask speaks, then waits: a bubble appears in the corner with the transcript and a reply box, and whatever is typed comes back on stdout.

answer=$(saynow --from "billing · migration" --ask "Drop the old table?")

Exit 0 means they answered, 2 means they dismissed it or were away — never read 2 as consent. --file report.md renders a Markdown document in the same bubble: headings, tables, code, links and images, local or remote.

The bubble needs a window shell — Xcode command line tools on macOS, or any Chromium-based browser. Without one saynow still speaks, --ask exits 2, and --file is read aloud rather than shown.

Both builds read the same config file, so installing the npm one alongside this changes nothing you have already set up.

Claude Code users can install the plugin, which teaches Claude when to speak:

/plugin marketplace add dhruvyad/saynow
/plugin install saynow@saynow

Full reference

saynow --help    # complete reference: every flag, setting, and exit code

License

MIT — source at https://github.com/dhruvyad/saynow