1 Assistant hour included freeStart free
1THING

Legal

Service Providers

Last updated: October 22, 2026

Summary

Your assistant uses five specialist AI services to turn speech into text and to write your notes, drafts and answers. We send each one the data its job needs, and the list below says what that is.

All five run in the United States, so US law applies while they hold your data. Anthropic may also process it in other countries where it operates, and Google may briefly store or cache it in other countries.

We do not use your calls to train AI models, and we never sell your data.

1. The Services We Use

This is the current list. Each entry says what the service does and what we send it.

Soniox

Turns the recorded speech of your calls into text, and your dictation into text as you speak.

What it receives: The audio of your calls, one side of the call at a time, and your voice while you dictate.

Where it runs: United States

Training on your data: Soniox does not use what we send it, or what it returns, to train or improve its models.

How long it keeps it: We delete each call recording and its transcript from Soniox as soon as the transcript comes back. Soniox does not store dictation: it processes it as it arrives.

Groq

Turns recorded speech into text: your meetings held in a language other than English, your English meetings if AssemblyAI is unavailable, and your calls when Soniox is unavailable.

What it receives: The audio of those meetings, of your calls when it stands in for Soniox, and the voice clips you dictate.

Where it runs: United States

Training on your data: Groq is not permitted to use what we send or what it returns to train or fine-tune any AI model.

How long it keeps it: Nothing. We have switched Zero Data Retention on for our Groq account, so neither the audio we send nor the text it returns is stored once the request is answered. That is a setting we turned on, not Groq's default: without it, Groq may log what we send and what it returns for up to 30 days.

Google Gemini

Builds the index that lets you search what was said across your transcripts. It also stands in on the follow-up card when Claude and OpenAI are both unavailable.

What it receives: Transcripts, not the audio.

Where it runs: United States, and Google may cache data in other countries

Training on your data: We use Google's paid service, and under its terms Google does not use the prompts we send or the responses it returns to improve its products. On the free service it would, so we do not use the free service.

Anthropic Claude

Writes the summary of a call or meeting, drafts the follow-up card (the suggested notes, tasks and CRM updates you confirm or discard), answers the questions you ask the assistant, drafts your emails, reads the business cards you photograph, and reads screenshots you attach to a support request so our support team can see what went wrong.

What it receives: Transcripts, the questions you ask, the emails you draft with the assistant, business-card photographs, and screenshots attached to support requests. Not the call audio.

Where it runs: Stored in the United States; processing may run in other countries

Training on your data: Anthropic may not train its models on what we send through the service.

How long it keeps it: Anthropic deletes what we send within 30 days. If its automated safety systems flag a request, Anthropic may keep that request and its response for up to 2 years, and its safety scores for up to 7 years, and it may keep data longer where the law requires. Anthropic may look at what it keeps for safety and security purposes.

OpenAI

Drafts the follow-up card when Claude is unavailable: the suggested notes, tasks and CRM updates you confirm or discard.

What it receives: Transcripts, not the audio.

Where it runs: United States; processing may run in other countries

Training on your data: OpenAI does not train its models on what we send through its API.

How long it keeps it: OpenAI may keep what we send, and what it returns, for up to 30 days to detect abuse, and longer where the law requires.

AssemblyAI

Transcribes your meetings held in English, and separates voices inside a single recording in any language: an in-person meeting, where one microphone picks up everybody, a virtual meeting with two or more people on the other side, where one channel carries all of them, and a call where a third person joins, on the side of the call that carries more than one voice.

What it receives: The audio of your English meetings, of meetings where voices need separating, and that side of those calls.

Where it runs: United States

Training on your data: AssemblyAI's standard terms let them train on customer data, so we have opted our account out. While that opt-out is on, they do not train on anything we send them.

How long it keeps it: We have set AssemblyAI to keep nothing for longer than one day. After that, deletion of both the audio we send and the transcript it returns begins. Their standard periods are longer than this.

Security: AssemblyAI holds SOC 2 Type 2 certification, audited every year, and has its systems tested by outside security firms.

2. When This List Changes

The list above is the one in force today. If we add a service, remove one, or change what one of them does, we update this page, and the date at the top tells you when it last changed.

3. The Rest of Our Providers

This page covers the AI services that read your calls, transcripts and documents. The full list of every service provider behind 1THING, including hosting, the database, payments, email and telephony, is in section 10 of our Privacy Policy.

Questions about any of this go to privacy@1thing.ca.