The Core Idea: Execution, Not Suggestion
Mica works by executing your Excel task directly in the workbook on your own Windows machine. You describe what you want, Mica shows the planned steps, you authorize them, and it runs each one while displaying the formulas and results. No workbook is uploaded — local execution is the design, not a promise.
That makes Mica an action-bot: AI that does the work and shows you every step. For the one-paragraph version of the product and the company behind it, start with what Mica is.
The same distinction runs through the whole category. Where a cloud add-in hands you a formula to paste — or executes on its servers with your file uploaded — Mica runs the change in your file on your machine. That axis — where the work runs and where the data goes — drives the execution vs advice comparison of the major tools.
A prompt like “average the last 12 months of revenue” gets different treatment depending on the tool: an advisor returns a formula or a number in a chat box, while Mica returns the formula written into your sheet, calculated against your rows. Both are helpful; only a local agent leaves you with a finished workbook on your machine with no workbook uploaded.
How a Task Runs, Step by Step
- You describe the task in plain language — for example, “remove duplicates in column A and total the sales by region.”
- Mica reads your workbook and turns the request into a plan of concrete spreadsheet operations.
- The plan is shown to you as a list of steps, each with the cells and actions it will touch.
- You authorize the plan, and Mica executes each step in your workbook one by one.
- The formulas and results for every step are displayed as it runs, so you can check the work as it happens.
Concretely, a request to “sum sales by region and add a totals row” becomes a plan that names the SUMIFS to write, the range it will read, and the cell where the totals row lands. You approve it, the formula is written into the right cells, and the result appears with the formula still visible — nothing pasted, nothing guessed.
The step-by-step plan also makes errors cheap to find. If a step reads the wrong range or writes to the wrong cell, you see it in the plan before it runs — or in the executed formulas afterward — and the fix is a rephrase or a single adjustment, not a hunt through a pasted formula you did not write.
The plan granularity is fine enough to matter: each step names the range and the operation, so a ten-step task reads as ten checkable lines rather than one opaque instruction. That is what turns approval from a formality into an actual review, and what makes larger tasks feel safe to delegate.
To see it from install to first execution, the how to use Mica guide runs through a complete example.
Permission & Verification
Every task runs through a permission step, so nothing changes in your file without your approval. The plan is presented before any cell is touched, and you can reject or adjust it.
Verification is built into the flow rather than bolted on. Because each step shows its formulas and results, you are not trusting a final answer — you are watching the operations that produced it, which is what makes the tool safe for files you cannot afford to break.
Rejecting a plan changes nothing — no step runs and your file stays untouched. You can rephrase the request and have it re-planned, which makes the permission step a working checkpoint rather than a formality.
Permission is per-task, not a blanket grant. Every request produces a fresh plan for your review, so one misphrased prompt cannot quietly trigger a chain of unrelated changes — the checkpoint resets each time, which is the difference between supervised and trusted automation.
What Stays Local
Mica processes your workbook on your own device. No workbook file is uploaded, no subscription is required, and no data is pooled for training — local execution is the architecture, not a feature to enable.
That is a category-level choice, not just a Mica one. For how cloud tools differ — what they upload, store, and may train on — the data upload vs local explainer compares the approaches side by side.
Local also means the file itself stays in place. Mica reads and writes the workbook where it lives on disk, so there is no exported copy sitting in a staging folder or a cloud sync queue — the workbook you close is the workbook that was worked on.
That local design shows up most clearly in how it treats your file. Your workbook and its data stay on the machine instead of travelling to a server, so client sites, corporate networks, and unstable connections change the tool's behavior far less than they would a cloud add-in that must upload the whole file. (The model conversation still needs a connection, as with any AI — what stays local is the workbook and its execution.)
Learn While Doing
Because every step is visible, using Mica teaches you the formula as it runs. When you watch a SUMIFS or an INDEX-MATCH get written against your own data, the syntax stops being abstract.
That learning-by-doing loop is the reason the product shows steps at all. Every reviewed step builds transferable Excel skill instead of a dependency on the tool.
A nested IF or an INDEX-MATCH that you watch being assembled against your own columns sticks in a way a copied answer never does — the structure of the formula is visible in the exact cells it runs on.
The loop pays off differently by skill level. For a beginner it turns the first weeks of formulas from memorizing into watching; for an experienced analyst it is a fast path to a formula you would have written yourself — either way, the visible steps convert the tool from a dependency into a teacher.