Knowledge base/How Mitta decides
How the coach answers, and what “sources” means
In short
The coach is an AI (Claude, made by Anthropic), not a person. It answers from a summary of your own training, food and weight, and it can look up more when it needs to. When it looks something up in your data, small chips under the answer show where the numbers came from. Tap the chip to open that workout, day or weigh-in and check it yourself.
What is the coach?
The coach is an AI model, Claude by Anthropic. The chat says so at the top (“AI coach · Claude”) and at the bottom: you are talking to an AI, not a person, and it can make mistakes. It is not medical advice.
It stays with training, food, sleep and recovery. Ask it something else and it says that is not what it is for. Answers are meant to be short: usually 2 to 5 sentences.
What does the coach know about you?
Only what the app gives it. With each question it gets:
- A summary of your profile: goal, experience level, training days, equipment, injuries you listed and your food targets. Your name is not sent, and your age is rounded to a 5-year band.
- Your current plan and training block.
- Your training per week and your last sessions, your food for the past week and your weight over the last 4 weeks.
- If you turned on the coach’s memory (a separate switch): the short facts you confirmed, such as which days you train.
When the coach needs more detail, it can ask the server for it, for example your workouts of the last 4 weeks or the progress of one exercise. It gets at most 6 of these look-ups per question. If something is missing from your data, it is told to say so instead of guessing.
What does “sources” mean?
It does not mean studies. It means where a number came from in your own data. When the coach reads something from your training, food, weight or plan, the app keeps track of which workout, day or weigh-in it was. The coach is told to mark each number it takes from there, and the marks turn into small chips under its answer:
- from workout 12 Oct: a single workout
- from training week, from food and from weigh-in, each with a date
- from exercise and from your plan
Tap a chip and it opens that workout, that day in your food diary, or your body progress, so you can see the number in its place. An answer shows up to 6 chips, and two chips that point to the same place are merged. Marks that do not match anything the coach looked up in that same question are dropped.
Chips appear when the answer used something the coach looked up. An answer without chips came from the summary, from general knowledge, or from your own question.
What is “Why?” under a proposal?
When the coach proposes a plan, a meal to log or a correction, the card has a Why? link. It opens two things:
- Reasons: up to 3 short lines the coach wrote, tied to your numbers (for example how many sessions you completed in the last 6 weeks). These are written by the AI. If your data is missing, it is told to say that in a reason.
- Based on: a list of what the coach was given, such as “4 weigh-ins in the last 4 weeks” or “no food logged last week”. The app builds this list in code from your data. The AI does not write it.
Nothing changes until you tap. The coach only proposes, and the card has a button to use it and one to dismiss it. A plan, a meal and a correction each wait for your tap, and so does a fact to remember.
What the coach cannot do, and how to check it
- It can be wrong. The server compares the numbers in an answer with the data the coach received, but only to keep a count. It never blocks or corrects an answer. Use the chips and check numbers that matter.
- It does not diagnose or treat. For pain that lasts, is sharp or spreads, dizziness, chest pain or shortness of breath, it tells you to stop and see a doctor or physiotherapist. When your message contains a possible alarm sign, the app shows a fixed card above the answer, and the coach gives no advice about the complaint.
- It keeps to fixed food limits. It does not advise under 1,500 kcal a day for men or 1,200 for women, crash diets or losing more than 1% of body weight a week. If you show signs of an eating disorder, it does not go along with a goal to lose weight and points you to a doctor.
- It does not set your lift targets. Those come from fixed rules. How Mitta sets your lift target
To check an answer: open a chip, compare the number with the screen it leads to, and tap Why? on a proposal. If it is wrong, use the thumbs under the answer and pick a reason (wrong, too general, tone or other). Under a Why? from the app itself, Ask more takes the topic into the chat.
How many questions do I get?
The coach is part of Pro + AI. That plan has 100 questions a month, with at most 15 a day. Invited testers have lower limits: 40 a month and 8 a day. The chat shows how many are left, and tells you when the limit is reached. Prices are not set yet. What each plan includes
You can turn the AI coach off in Settings. Then the app works as before, without the coach.

How Mitta does it
Answers you can check.
The coach works from your own numbers and shows where they came from. Proposals come with a reason, and nothing changes until you say yes.
- Chips under the answerTap one to open the workout, day or weigh-in behind a number.
- Why? on proposalsReasons with your numbers, plus a list of what the coach was given.
- You decidePlans, meals, corrections and facts to remember all wait for your tap.
- Fixed limitsFood floors and a safety card for alarm signs, set in the app, not by the AI.
Ask, then check.
The coach answers from your own data and shows where a number came from. Free to start, in the web app for invited testers.
Questions about
How the coach answers, and what “sources” means
Are the sources scientific studies?
No. A source is a place in your own data: a workout, a food day, a weigh-in, an exercise or your plan. Where research matters, the knowledge base articles list their studies.
Why does an answer have no chips?
The coach did not look anything up for it, or it used the summary and general knowledge. Ask it to check a specific week and it can look it up.
Can the coach change my plan or log food by itself?
No. It proposes, and you tap to use it. It cannot save a proposal for you.
Does the coach remember our chats?
Your chat stays until you delete it (New conversation, then Delete chat history). What the coach remembers about you across chats is only what you confirmed, if you turned memory on. You manage that in Settings.
Is my name sent to the AI?
No name or email address is sent. A summary of your data, your messages and meal or label photos go to Anthropic when you use the coach.
Sources
Written by Cody de Leeuw. First published 27 October 2026, last checked 27 October 2026. This article is general information, not medical advice. Spotted a mistake? Tell me