Voice intake platform

Voice intake that costs nothing per response, in your customer's language.

Describe the conversation you want once. SpeakingForm runs it as a web form, an embed, a voice note, or a live voice agent — and answers your customer's questions from your own documents while it's at it.

Live form — try it

Built for custom-order shops, field-services teams, and the agencies that serve them.

Watch a sentence become a form.

Your customer talks the way they'd talk to a person. SpeakingForm turns it into clean, labeled, redacted fields — the ones you'd have typed by hand.

Spoken words in amber. Captured data in teal. That's the whole system — sound in, structure out.

The economics

No per-response meter, by design.

SpeakingForm is built to run on self-hosted inference over a stateless GPU-plus-Postgres stack that reuses almost every prompt prefix. The marginal cost of one more response is engineered toward zero — so pricing doesn't punish you for succeeding.

How the architecture works

Privacy, upstream

The model never sees raw PII.

Names, phone numbers, addresses, and card-shaped strings are pulled into a vault and replaced with tokens before any language model reads the text — then rehydrated only into the record you own. Pin the vault to an EU jurisdiction when residency is a requirement.

Read the security posture

One configuration

Author once. Render everywhere.

Describe the interview once. SpeakingForm renders it as a web form, an embeddable widget, an async voice note, or a live voice agent — and you can switch or mix channels per form without re-authoring a thing.

See every channel

It answers back

Your form knows your business.

Upload your menus, policies, or price lists. When a respondent asks a question, the form answers from your documents — grounded with a citation — and refuses cleanly when the answer isn't in there. No hallucinated promises.

How grounding works

Do you do vegan buttercream?

Yes — we make a coconut-oil vegan buttercream on request, with no dairy or egg.

from: Allergens & substitutions, p.2

Four channels. One brain.

Web form

A clean, guided web form on your domain.

Shipped

Embed widget

One script tag; renders anywhere; isolated styling.

Shipped

Async voice

Customers record; you get structured data back. In-browser today; WhatsApp voice notes are on the roadmap.

Shipped

Landing

Live voice agent

Full-duplex conversation with barge-in and backchannel acks.

On the roadmap.

WhatsApp & SMS voice notes

We're building intake over the channels your customers already message on.

Roadmap

Multilingual rollout

We're rolling out more languages — the engine is language-agnostic by design.

Roadmap

Premium cloned voices

We're adding premium, brand-matched voices for live and async replies.

Roadmap

Talk for three minutes. Leave with a form.

Describe what you need out loud. We turn it into a working intake form.