---
title: "No AI in TaskPocket — and why that is deliberate"
description: "No assistant, no summariser, no model reading your list. The date parser is a grammar, not an LLM. What that buys you, and what it costs you."
canonical: https://taskpocket.io/no-ai/
lastUpdated: 2026-08-26
---
1. [Home](https://taskpocket.io/)
2. No AI

# There is no AI in TaskPocket

There is no AI in TaskPocket. No assistant, no summariser, no model reading your list. The natural-language date parser is a deterministic grammar rather than a language model — no round-trip, no quota, no guessing — and nothing you write is ever used as training data by anyone.

By [Bohdan Stefaniuk](https://taskpocket.io/about/) Last updated 2026-08-26

## What “no AI” means here, precisely

The phrase is doing a lot of work on a lot of landing pages, so here it is as four claims you can check rather than a mood.

1. The date parser is a grammar, not a model. It is built on chrono-node plus a hand-written rule set for recurrence. “Call the dentist tomorrow at 9” is matched, not inferred.
2. No task text leaves the service. Parsing happens without a network round-trip, so it works with the wifi off and there is no quota to run out of.
3. Nothing you write is training data. Not for us, not for anyone we buy services from.
4. There is no assistant of ours reading your list. Nothing summarises your week, suggests priorities, or drafts anything in the background.

## The objection is not to AI. It is to AI you did not choose

Almost nobody who searches for a to-do app without AI is against the technology. They use it every day somewhere else. What they object to is narrower and more specific, and it is worth separating out, because conflating the two is how this position turns into posturing.

The complaint is about AI that arrived: switched on by default, in an app they chose for being small, operating on the most private list they keep, with a model they did not pick, and — often enough — landing in the same release as a price rise. None of that is an argument about whether language models are useful. It is an argument about consent and about who is choosing.

So the position here is not that AI has no place near a task list. It is that it should be yours, chosen by you, and off until you turn it on. Which means none of it in the product, and none of it in the price.

## Why this became a question in 2026

It is a fair thing to be sceptical about, so rather than characterise anyone's users, here is what the vendors themselves publish.

Todoist lists four AI features on its pricing page — Filter Assist, Task Assist, Email Assist and Ramble — with Ramble available on the free tier in limited sessions. TickTick sells “AI Features” as a Premium-only row in its own free-versus-Premium comparison table, alongside custom filters and integrations.

Both are reasonable products making a reasonable bet. The point is only that the bet is now the category default, which is precisely why somebody typing “to-do app without AI” is having trouble finding one. If you want the other option, it has to be somebody's deliberate choice, and this is that choice.

## Deterministic beats probabilistic for this particular job

There is a real engineering argument underneath the positioning, and it is not “models are bad”.

For reading a date out of a sentence, a parser that is right 97% of the time is worse than one that is right 100% of the time on a narrower grammar — because you cannot tell which 3% you are in. A model that quietly misreads “first Monday” produces a task on the wrong day that looks exactly like a task on the right day. A grammar either matches or does not, and when it does not it says so and leaves your text alone.

The narrower approach also has properties the probabilistic one structurally cannot: it behaves identically every time, it works offline, it costs nothing to run, and it cannot be the reason a price goes up. For a field this well-defined, that trade is not close.

## If you do want an assistant

Then it should be your model, on your terms — you decide which one, you decide what it can see, and it does nothing until you connect it. That is the position, and it is a statement about what will never be built into the product rather than a promise about what will: there is no way to connect a model to TaskPocket today, and this page is not announcing one.

Connecting your own model is not a novel idea and would not be sold as one. TickTick already runs a first-party MCP server, and Todoist documents connecting ChatGPT and Claude Code to its own. The thing that is actually different here is what stays out of the product: no assistant of ours, reading your list, by default.

## Where AI would genuinely help, and why it still stays out

A page like this is worth nothing if it will not name the cases where the other side is right. There are several.

Turning a paragraph of meeting notes into six tasks is real work that a model does well and a grammar cannot do at all. So is triaging a backlog you have let rot for three months, or answering “what did I actually get done this quarter”. People who use those features are not being sold something useless — they are getting something this app does not offer.

It stays out because the cost is not visible in the feature list. An assistant needs to read everything to be useful, which makes the privacy promise conditional. It needs a server round-trip, which makes the app slower and dependent on somebody else's uptime. It needs paying for, which shows up in your subscription whether or not you use it. And it makes the product bigger, which is the one thing this product is trying not to be.

That is a choice for a segment, not a claim that the segment is right and everyone else is wrong.

## What this costs you

Plainly, because you would find out in a week anyway.

- No natural-language anything beyond dates and repeats. The parser reads “every 2 weeks” and “tomorrow at 9”. It will not turn a paragraph into a project.
- No smart triage, no suggested priorities, no weekly summary. Reviewing your list is a thing you do, and the app gives you views to do it with rather than doing it for you.
- The parser has a vocabulary, and you can hit its edge. Unusual phrasings are left as plain text rather than guessed at — correct behaviour, and occasionally an annoying one.
- There is nothing to connect a model to. TaskPocket has no public API and no MCP server, so “bring your own model” is a design principle rather than something you can do this afternoon.

## Sources

1. [Todoist — Pricing (the four named AI features, and Ramble on the free tier)](https://www.todoist.com/pricing) checked 2026-08-23
2. [TickTick — Upgrade ("AI Features" as a row in their own plan comparison)](https://ticktick.com/upgrade) checked 2026-08-23
3. [TickTick — TickTick MCP (their own first-party MCP server)](https://help.ticktick.com/articles/7438129581631995904) checked 2026-08-26
4. [Todoist — Use ChatGPT with Todoist (MCP)](https://www.todoist.com/help/todoist/todoist-and-ai/use-chatgpt-with-todoist-mcp-WEeLx9d8h) checked 2026-08-26
5. [chrono-node — the date-parsing library the deterministic parser is built on](https://github.com/wanasit/chrono) checked 2026-08-26

No assistant, no quota, no round-trip. The free plan has unlimited tasks and needs no card.

Written by [Bohdan Stefaniuk](https://taskpocket.io/about/), who builds TaskPocket. If you disagree with any of this, the [contact address](https://taskpocket.io/support/) is a real inbox.

## Related reading

- [Recurring tasks that survive real life](https://taskpocket.io/features/recurring-tasks/) — Nth weekday, the last business day, stop after ten, skip one, and repeats counted from the day you finish. What each app can express, with sources.
- [Start dates and deadlines — two dates, two questions](https://taskpocket.io/features/start-dates-and-deadlines/) — When you plan to start and when it is actually due are different questions. Here is how the two dates work in TaskPocket, and which apps have both.
- [Todoist vs TaskPocket (2026) — an honest comparison](https://taskpocket.io/compare/taskpocket-vs-todoist/) — Todoist is bigger, has reminders and runs everywhere. TaskPocket is smaller, has no AI and no subscription. Where each one actually wins, with sources.
- [Deadline vs due date: which one does what](https://taskpocket.io/guides/deadline-vs-due-date/) — A due date schedules work; a deadline records a consequence. Here is what each one is for, a rule for deciding, and which task apps give you both.
