TaskWaffle AI Disclosure

Version 1.2, amended 24 September 2026 (the companion's daily lines), 1 October 2026 (a weekly free allowance, more generous) 17 October 2026 (Groq's training and retention terms, now reviewed) and 2 November 2026 (voice typing). Published 16 September 2026. Effective 16 September 2026.

This notice explains how TaskWaffle processes information when you use its AI assistant.

The assistant is useful because it can see your real data. That deserves a plain explanation of exactly what happens.

What gets sent, and when

To the AI provider. Something is sent to Groq when you send the assistant a message, or press a labelled AI button such as "Break this down with AI" or "Ask your week". Opening TaskWaffle does not start an AI request. Nothing runs on a timer, in the background, or while you are just moving around the app.

One exception, which you switch on yourself. On the Plus plan, the companion character on your Home screen can say a short line now and then. If you turn that on (it asks you once, and the switch is in Settings under Appearance), TaskWaffle sends one request a day to Groq asking for those lines, built from the same modules you have chosen for the assistant and nothing else. The companion never comments on your journal. Turn the switch off and this request is never made.

Everything else is ordinary app traffic. Signing in, syncing your records, loading the app, buying a subscription and delivering a reminder all involve our servers and our providers, and none of it goes to Groq. The Privacy Policy covers that side.

Nothing is readable until you say which parts are

The first time you use the assistant, it stops and asks. That screen names Groq, says that what it reads and what you type are sent there, and lists every module with its own tick box.

Every box starts empty, including ordinary ones like Tasks and Calendar. A box we ticked for you would not be your answer. Bills and subscriptions count as part of Money.

The same screen asks separately whether you are 18 or over, which Groq's terms require. TaskWaffle itself is for 16 and over, and everything except the assistant works either way. Nothing is sent until both questions are answered and stored.

A module you have not ticked is not read: not summarised, not anonymised, not sent. Settings, then AI assistant, is where you change any of it or withdraw the whole permission, which puts the account back to reading nothing.

Exactly what travels with a request

• Your message, as you typed it.
• The recent conversation, up to about thirty turns. A conversation is sent again in full with each new message, so something you said earlier travels again.
• Selected fields from the modules you ticked. Not whole records: named fields, with long free text shortened. Notes are cut to about 160 characters, a journal entry to 200, a review to 300. Dates, times, amounts, currencies and titles are sent as written.
• Facts you asked it to remember, if you ticked Memory.
• Your upcoming Google events, if you ticked Calendar and connected Google.

Two things travel whatever you tick, and this page did not say so before version 1.2:

• Today's date and your time zone offset. An assistant asked about "tomorrow" cannot answer without them, and taking them from the server would put everyone in the wrong day.
• The names of the modules you did not allow. The model is told it is working with less, so it says "I cannot see your money" instead of inventing a number. The names travel; nothing from inside those modules does.

Neither carries any of your content.

Before a request goes out, the assistant panel lists the modules it is allowed to draw on.

How the permission is enforced

Your permission is a stored record holding the version of this page you agreed to, the provider named on it, the modules you chose and your age answer.

It is enforced by the server, not by the app. The server reads that record itself and applies it to every request, and an account with no record has every request refused before anything is read and before anything is sent anywhere. A request may ask for *less* than the record allows, never more, so a modified copy of the app cannot widen its own access. The tools the assistant is offered are filtered too, so it is not told that a module it may not read exists.

When the module list, the provider, or what this page discloses changes, the version moves and you are asked again from a blank sheet rather than being held to an answer you gave about something else. A module added to the app after you answered stays unavailable to the assistant until you are asked again. You will never find a new module switched on for the assistant because we added one.

Each reply carries the identifier of the policy that produced it, and that decision is recorded, by module name only.

What happened on 16 September 2026. Three modules were removed from this list because the pages behind them were removed from the app: Contacts, Documents and the Emergency card. All three existed in the code and none of them had ever been reachable from any screen, so the consent screen had been asking permission to read pages that did not work, including, in the emergency card's case, blood group, allergies and next of kin. A shorter list is still a changed list, so every permission given against the old one is void and everyone is asked again.

An honest note about what came before. Until version 1.1 of this page, permissions were stored as a list of exclusions. That made "I have never opened this screen" indistinguishable from "I agree", and a module added later would have been readable without anybody agreeing to it. It is why every account is asked again from a blank sheet rather than having an old answer carried over.

There is one thing none of this protects you from, and it is worth saying plainly: it does not help if somebody is using your signed-in session, because they can simply switch the modules back on. The permission governs what leaves the account. Keeping other people out of the account is what your password and your device are for.

Who processes it

Replies are generated by Groq, Inc., a third-party AI inference provider, a separate company from TaskWaffle. Groq, Inc. is incorporated in Delaware with its principal place of business in California, and requests go to its United States endpoint, api.groq.com. We do not choose a region and cannot tell you which data centre serves a given request. Groq's own terms require users to be 18 or over, which is why the assistant asks your age separately from the 16+ declaration at sign-up.

Our account is on Groq's standard Services Agreement, which we reviewed on 17 October 2026. Section 4.2 says Groq may not use what we send, or its replies, to train or fine-tune any model unless we instruct it to, and we have not. Groq's documentation says it keeps no customer data from a request by default, except logs it may keep for up to 30 days to fix a fault or investigate abuse. We have not switched on Groq's zero-retention setting, so those logs can exist. There is no separately negotiated agreement; the standard one applies, and Groq can change it. TaskWaffle itself never uses your content to train AI models.

The underlying model may change as better options appear. If a change alters what is disclosed here, the version moves and you are asked again.

Voice typing

If you have allowed the assistant, including the 18+ question, the mic in Quick capture and Notes asks once whether to send your recordings to Groq to be turned into text by its Whisper speech model. If you agree, each clip (up to about three minutes) is sent from our server to Groq and the text comes back into the box you were typing in. We do not store the recording or the text on our server. Groq's terms are the same as for the assistant: it may not train on what we send, and it may keep fault or abuse logs for up to 30 days. Voice typing reads none of your modules. If you decline, or have not allowed the assistant, the mic uses your browser's speech service instead, and in Chrome that service is Google's.

What the assistant can and cannot do

It can read, subject to your ticks, and it can propose changes: a task, an event, a plan for your day. Every proposal shows you exactly what it wants to do, and nothing happens until you press Confirm. Confirmed changes are recorded server-side.

It cannot act on your data by itself, and it has no reach outside your own TaskWaffle records. It cannot send an email, invite anybody to anything, write to your Google Calendar, or share anything with another person.

Memory

If you ask it to remember something ("I pray at 6", "Fridays are for family"), that fact is saved only after you confirm it, is visible in Settings, then Assistant, and can be deleted there at any time.

Memory is a module like any other: it is not readable until you tick it, and when it is unticked the assistant is not offered the tools to save or forget anything either.

Deleting a saved fact also clears the open conversation. A conversation is sent back in full on every turn, so leaving it open would carry the deleted fact along with it.

Google Calendar

Connecting Google Calendar is optional. We ask Google only for read-only access to your events; the app makes no write calls and cannot create or change a Google event.

If you tick Calendar in the assistant permission, those events are part of what is sent to Groq. If you untick it, they are not.

Disconnecting deletes our stored tokens and clears the copied events from your device. We also ask Google to revoke the token, but Google does not confirm that to us, so check the third-party access list in your own Google account if you want certainty.

Honesty about limits

AI answers can be wrong, including the calculations and the dates. The assistant works from real numbers in your workspace and your device's clock, which makes it more useful than a chatbot guessing, but it does not make it correct. Treat anything important as a draft, not a verdict, and check it before you act on it.

Reporting a reply. Every reply has a Report button under it. Use it for anything wrong, inappropriate or unsafe. The reply and your note are sent to TaskWaffle under your account so it can be looked at; nothing else from the conversation is. You can also write to contact@taskwaffle.com.

It is not a doctor, therapist, lawyer, accountant or financial adviser. It is not an emergency service, and a reminder should not be your only safeguard for medication, safety or a deadline that matters.

Limits and cost

Assistant use is capped per account, to prevent abuse and keep the service affordable. The free plan allows 21 questions over any 7 days plus a limit on the amount of text processed over the same 7 days; each question comes back 7 days after the day it was asked (days are counted in UTC, and the app shows the time in yours). Plus has no question count. It has a text allowance of 250,000 tokens a day (a token is roughly three quarters of a word, counting what you type, what the assistant reads and what it writes back), which is about 50 typical questions, and it renews at midnight UTC. The 14-day trial has 50,000 tokens a day. On every plan, at most 30 questions can be sent in any 5 minutes. Settings, under AI assistant, shows how much is used, how much is left and when it comes back, in your own time. A request that fails does not count against you. When you reach a limit the app says so plainly.