The FoodProt Assistant

Opening it and choosing what to ask about

Open the assistant with the "Ask FoodProt AI" button on any page when you are signed in, or from Tools → Assistant inside a project. You then choose one of two modes.

• General help: answers come from these help pages. No project data is read.
• A project you choose: answers come only from that one project, and only from what your role allows you to read. Projects are never mixed.

How answers are labelled

Each answer is labelled by kind, so you can tell a count from an explanation. When the assistant has no evidence, it says so instead of guessing. It can still be wrong, so open the linked evidence before relying on an answer.

• Database result: counts and lists computed from the project’s records.
• Document evidence: quoted passages with the paper title, DOI, page, table or figure, and an "Open evidence" link.
• Platform help: an answer taken from these help pages.
• Model-assisted explanation: shown only when the platform has this switched on, and always tied to quoted evidence.

Questions about the evidence

Inside a project you can ask about the values and documents themselves. The assistant answers from this project’s verified values, extracted values and the documents that have been read — never from outside knowledge. If nothing matches, it says so.

• Which papers report something: "Which papers report pea protein hardness after heat treatment?" Only values a reviewer has approved are counted.
• Values with conditions: "What are the water holding capacity values at pH 7?"
• A comparison: "Create a comparison table for pea and soy protein solubility." Each unit gets its own row; values in different units are never combined, and the table is not a statistical test.
• A table you can save: "Create a CSV of gel hardness values." Download CSV is offered if your role includes exporting, and each download is recorded in the project’s activity.
• A summary of stored values: "Summarise the evidence for ultrasound." It lists counts and ranges, not conclusions.
• Where a value came from: open the value, then ask "What is the source for this value?"
• Why a paper was excluded: open the paper, or give its DOI. You get what was recorded, or a plain statement that no reason was recorded.
• What a paper says: open it in the reader and ask "Summarise this paper". The answer quotes the paper’s own opening sentences; it does not rewrite them.
• Selected papers: tick papers in the Library, then ask "Summarise the selected papers" or "Draft a paragraph from the selected papers". The draft restates verified values only, for you to edit.
• Support for a statement: "Find passages that support this statement: …" returns passages to read. The assistant cannot judge whether they support it, and never suggests a reference that is not in your library.
• Long lists end with "Show the next …", which continues the same list.

Actions need your confirmation

The assistant can propose a small set of actions. Each one is shown to you first, runs only after you press Confirm, and runs only if your role allows it.

• Build an export.
• Find full text.
• Start extraction.
• Run quality checks.
• Retry a failed background task.

What it never does

Some things are outside what the assistant can do, whoever asks.

• It never deletes anything.
• It never changes roles, the team or settings.
• It never approves or rejects a value.
• It never shows passwords, keys or other credentials.
• It treats text inside papers as content to quote, never as instructions to follow.
• It never writes into a manuscript or a workbook. Citations are inserted by you, in the Writer.
• It never answers from general knowledge or the web. Its only sources are this project and these help pages.

Your conversations

Conversations are private to you and kept separately for each project. You can clear them at any time, and they are removed automatically after 90 days.

The assistant is most useful for looking things up across many records, such as "Which papers still need a PDF?" or "Which values could not be standardised?". Decisions such as whether a paper is in scope remain yours.