An offline voice pack that does the work

Buy a pack once, from $24 to $199. Besides the audio it includes a small agent that uses the pack's own lines to message your customers, answer the phone and email, tell your team on Slack or Teams, and keep your calendar in step. It runs on your own server, so there is no Speakvora per-use fee.

What one pack does

  • Confirms bookings and orders by text, and adds reservations and appointments to your Google Calendar.
  • Reminds customers a day and a couple of hours before, straight from your Google or Outlook calendar.
  • Tells people their order is ready, running late, or their table is ready, by text, WhatsApp or a phone call that plays the pack's voice.
  • Answers texts such as “what are your hours?” and “where are you?” with your real hours and address, and passes “I want a person” to your team.
  • Answers your phone with a menu in the pack's voice: hours, booking, order status, or a person.
  • Answers email: replies to common questions, creates a ticket in your helpdesk, and alerts your team when someone needs a human.
  • Tells your team on Slack or Teams, and lets staff text a customer a pack line with a Slack command.
  • Follows up after a no-show, and thanks customers with a review link after a visit.
  • Posts a morning briefing of today's bookings.
  • Wakes your on-call engineer by text and phone call when a GitHub or GitLab pipeline fails or a PagerDuty incident opens, and posts it to Slack or Teams.
  • Protects your customers: texts only with consent, STOP is honoured, quiet hours, per-person limits, and a kill switch.

Connect whatever you use, once

Ready-made mappings exist for GitHub, GitLab, PagerDuty, Stripe, Shopify and Calendly, and it works with your existing Twilio number or any other provider's web API for texts and calls. Any system that can send a webhook (Square, Toast, a booking plugin, your own app) is connected by a short mapping in one settings file, with no code. Any system with a web API can be called from a workflow. Your own app only needs to send one signed web request when something happens, and it never waits for the agent, so a problem in the agent can not break your app. The agent is a separate program: it never touches your database.

A restaurant’s day, from our test run

  • A guest reserves a table: they get a confirmation text, the booking appears in Google Calendar and the host channel on Slack.
  • A day later the calendar triggers a reminder text. Two hours before, a second one.
  • The kitchen taps Ready: the customer who agreed to texts gets “Good news, your order is ready for pickup”. A customer who did not agree gets no text, and Slack shows that.
  • Someone texts “STOP”: they are unsubscribed and every later message to them is blocked.
  • Someone calls: the pack's voice offers hours, reservations, order status or a person.

This run used simulated Twilio, Slack and Google services on one machine, more than 60 automated tests, and our fake restaurant app. Start in dry-run with your own accounts: the agent records what it would do and sends nothing until you switch it on.

Guides

If a line is missing

Workflows ask for named moments (“order ready”), not phrase numbers. If your pack lacks one, the agent says so and can send a request to Speakvora: our phrase bot writes the line natively in your language, checks it, voices it, and adds it to every pack of that industry and language in the next version.

What it costs

The pack price is a one-time fee and includes the agent. We charge nothing per message or per call. Your own providers bill you directly on their terms: Twilio for texts, calls and WhatsApp, and Slack, Google or Microsoft for their services (many are free for small use).

Honest limits

  • It runs on your own computer or server (Python 3.9 or newer). Setup takes a developer or a patient owner about 15 minutes with the included guide.
  • US texting needs your Twilio number registered with carriers. Phone calls need a public address for the agent; the pack's .m4a lines need ffmpeg installed to play on calls (the workflow lines are mp3).
  • It answers by keywords and menus. It is not a conversational assistant; for open questions use our voice agent API.
  • Voices are synthetic. Non-English lines are machine-checked and native-review pending. Not for medical advice, emergencies or patient information.
  • It is built and tested against simulated services. Run it in dry-run against your own accounts first.
Browse offline packsDeveloper docs