Does my data go into model training?

No. We use the Lovable AI Gateway, which does not pass prompts to third-party providers for training. Details in the privacy policy.

Which model runs under the hood?

By default a lightweight Gemini through the Lovable AI Gateway, picked per question type. We will swap it when a better one appears.

Does the coach remember our history?

Within a single chat session — yes. Between sessions — no, on purpose. Every question starts fresh with current data.

What if the coach proposes nonsense?

Ignore it. You accept every suggestion manually, nothing lands in tasks without a click. It is a suggestion tool, not automation.

AI productivity coach — without the slogans

An AI coach sounds like a buzzword and often is one. In TaskAura the coach is a language model wired to your data: tasks, pomodoro sessions, habits, weekly plans. It answers specific questions (where do I lose time?), not existential ones (how do I become a better person?). That second one is for people, not models.

How it works

1
The coach only sees what you show it
Daily brief, today's tasks, the last week's stats. Nothing leaks outside your account.
2
Ask a concrete question
Why am I not finishing Friday tasks? Which habits do I miss most often? Suggest a plan for tomorrow considering my working hours.
3
Verify the answer
The model hallucinates. Every number in the answer should be verifiable in the app. The coach is a second brain, not an oracle.
4
Come back after a week
The weekly review asks about trends. That is where AI has the most to add, because it sees a pattern you do not remember.

Why it works

Externalising self-reflection (an outside interlocutor) is one of the better studied coaching mechanisms — Grant (2014) shows that simply asking structured questions improves self-regulation. A language model does not replace a coach but gives access to those questions 24/7 at a fraction of the cost.

The second argument is data. A human coach remembers what you said last week; a model has your actual behaviour in context. Subjective “I am working hard” gets measured by objective “last week 11 pomodoros, 7 of them finished.”

Sources

Grant, A. M. (2014). The Efficacy of Executive Coaching in Times of Organisational Change, Journal of Change Management, 14(2).
Ji, Z. et al. (2023). Survey of hallucination in natural language generation, ACM Computing Surveys, 55(12).
Bender, E. M. et al. (2021). On the Dangers of Stochastic Parrots, FAccT 2021.

When this method does NOT work

Language models hallucinate numbers and dates — especially risky in the “how much did I do last week” context. Always verify in the stats view. The coach is not a therapist either: if you are dealing with burnout, depression or anxiety, AI does not replace a specialist. It will help you plan, not heal.

In TaskAura

The coach is a separate view with chat, with access only to your data on this account (not other users', not the internet). Data sent to the model is minimal — passed in fragments only when you ask a question. You can turn it off entirely in settings.

Try it free

Frequently asked questions

Does my data go into model training?: No. We use the Lovable AI Gateway, which does not pass prompts to third-party providers for training. Details in the privacy policy.
Which model runs under the hood?: By default a lightweight Gemini through the Lovable AI Gateway, picked per question type. We will swap it when a better one appears.
Does the coach remember our history?: Within a single chat session — yes. Between sessions — no, on purpose. Every question starts fresh with current data.
What if the coach proposes nonsense?: Ignore it. You accept every suggestion manually, nothing lands in tasks without a click. It is a suggestion tool, not automation.