Most spreadsheet AI is a parlor trick: a formula generator that has never seen your file. The AI chat panel in RixSheets is wired into the workbook instead. It carries workbook tools that read the actual ranges, cells and formats on the sheet you have open — so its answers come from your data, its operation proposals fit your layout, and its summaries describe the numbers in front of you rather than a hypothetical. And it is not metered: AI pages in RixSheets are unlimited on every plan.
What the assistant can actually see
The panel's workbook tools give it structured sight of the open file:
- Ranges and cells — values, labels, empty blocks and the shape of the used area.
- Formulas — the real formula text in each cell, not just displayed results.
- Formats — number formats and styling, so it knows which column is currency and which is a date.
That context is the difference between "here is a generic SUMIFS example" and "here is the SUMIFS for your Units and Region columns, with the criteria range on your Lists sheet." It also means the assistant knows when the answer is already on the sheet — ask where the March total lives and it points at the cell instead of teaching you how to build one.
Reading: ask questions about the data
The first mode is reading. Ask in plain language:
- Which region has the lowest average margin, and how far behind is it?
- Are there any negative values in the Amount column?
- How many rows are missing a value in the Date column?
The assistant reads the relevant ranges and answers from what is actually there, usually pointing at the cells it checked. This is the fastest way to sanity-check a workbook you did not build — pair it with the audit moves from Name Manager and Go To for a full orientation.
Explaining: demystify formulas
Select any cell with a formula and ask what does this do. The assistant walks through each function and argument in order — what the range is, what the criteria mean, what happens at the edges. Nested formulas stop being archaeology. It works in the other direction too: describe the calculation you want and it drafts the formula, ready to paste or apply, with the formula engine guide covering everything it might reach for.
Proposing: operations you approve
The second mode is action. Ask for something — add a revenue column, convert units at the prices in column C, flag every order older than 30 days — and the assistant responds with a proposed operation, written as a structured description of the changes. The proposal shows:
- What it intends to write or calculate, cell by cell or column by column.
- Where the changes land — the exact target range.
- Why, in one line, so you can judge the logic.
You approve, refine the request, or reject it. Nothing touches the sheet until you say so, and the edit journal records what was applied — the trail is covered in Name Manager and Go To.
The proposals read like instructions to a careful assistant, and that is the right mental model: it is excellent at the mechanical, tedious, error-prone work — building a lookup column across a thousand rows, converting units at column-specific rates, flagging every overdue record — and it never tires of doing it the same way twice. The judgment stays yours: whether the threshold is 30 days, whether the flag means "call the customer" or "write it off", whether the column belongs there at all.
Summarizing: numbers into sentences
When a report or an email needs a paragraph, ask for one: summarize this sheet for the monthly update. The assistant reads the figures and drafts a summary grounded in the actual totals and trends — the pivot highlights, the outliers, the direction of travel. Edit the draft like any text; it arrives as a starting point, not gospel.
Summaries are also the cheapest way to test the panel's understanding. If its two-sentence take on the sheet matches your own read, its other answers deserve confidence; if it misreads a subtotal as a raw value or misses a filtered state, you have learned something about the sheet's structure worth fixing — and the fix usually makes the sheet clearer for human readers too.
Privacy and providers
The panel runs on whichever AI provider you configured in Settings — built-in Rix AI with your plan, your own API key, or local models via Ollama for fully offline analysis. Provider setup lives in AI provider setup, and the roster behind the panel is described in the AI agents overview. With unlimited pages, the honest way to use the panel is iteratively: ask, check its answer against a cell or two, ask again.
The habit that separates useful AI use from noise is verification by spot-check. When the assistant reports the lowest-margin region, click one cell and confirm the arithmetic by hand once; from then on you know how it reads the sheet and where its blind spots are. Every serious user of this panel has, at some point, discovered a structural surprise in their own workbook this way — a filtered row the summary missed, a subtotal mistaken for a value — and every one of those discoveries was a real problem the panel found for free.
What to explore next
- Formulas and functions — the engine the assistant drafts for.
- Getting started with RixSheets — open the panel in your first workbook.
- AI agents overview — the team behind the composer.
- AI credits explained — how unlimited pages fit the plan system.