How answers work

Two kinds of answer

A tool answer comes straight from a service built for the job: a weather service, an exchange-rate feed, a stock quote, a calculator. AnalyzeIt calls it and shows you what it returned, with the source. No AI model writes the number, so it cannot be invented. These answers carry the label From tools · 0 tokens, because no model tokens were used.

A model answer is written by an AI model. AnalyzeIt uses models for questions that need reading, comparing or explaining: your own files and records, a long document, or an open question. The model can call tools while it works, and when it does, the tool results are what it is allowed to state as fact.

What happens when a tool fails

If a tool cannot answer (a city it cannot find, a feed that is down), AnalyzeIt says so and does not fill the gap with a guess. You will not see the “From tools” label on an answer where the tool failed.

Your own data

Questions about records you have connected (for example “how many charges failed?”) are answered by querying the records, and the count comes from the database, not from the model counting rows. The assistant only reads a project’s data if you have allowed it in that project’s AI data access settings, and you can turn that off at any time.

Long documents

If you paste a long document and ask about it, Free keeps the most relevant excerpts and answers from those. On Premium and VIP, AnalyzeIt reads the document one section at a time: a small model notes what each section says about your question, and the answer is written from those notes. Premium reads about 150,000 characters this way and VIP about 300,000. If your document is longer, the most relevant sections (plus the first and last) are read and the answer says that parts were not read. You can watch each section being read as it happens.

Sources and citations

Answers from live tools and the web list their sources. Answers built from your stored notes and files show a From your notes chip that names the source and shows the passage used.

What a model answer still can get wrong

Models can misread or over-summarise. For anything that matters, open the cited source or the record itself. Where AnalyzeIt can check a figure against a tool or your data, it does; where it cannot, treat the answer as a draft.