What Data Mica Processes
- Uploads: no workbook file and no original dataset. The file itself never leaves your machine — the model conversation sends only the minimal task context it needs to plan, never the workbook, and no copy is stored or persisted on any server.
- Does not upload: workbook source files, full cell databases, macros, or private file assets — there is no account holding your files, no server storing your data, and no training pool receiving your business data. Where traditional cloud tools sync your entire workbook to their servers for processing, Mica cuts the exposure at the source: every data read and write happens in a local loop on your machine.
This page is the product-side statement. is it safe to upload Excel to AI covers the same question for the whole category, cloud tools included.
Local processing still means work is done on your data — reading ranges, evaluating formulas, deciding next steps. The difference is where that work happens: on-device, the execution is a function of your own machine, while in the cloud the same computation is a transfer of your data onto someone else's infrastructure.
Who can see your workbook after a task is simple to state: the local Windows user running Mica. There is no second copy on a server, no support tool with access, and no analytics feed that samples your cells — the file's audience is the machine it lives on.
For the one-paragraph version of the product, what Mica is is the place to start.
Local Execution Architecture
Local execution is the architecture, not a setting you enable: Mica executes tasks in your workbook on your device, so there is no cloud round-trip for the workbook file to travel. There is also no account that holds your files on a server, and no telemetry pool — properties that follow from the design rather than from a policy choice. A question that follows naturally is where the processing sits when the interface is a browser tab: it sits in the local engine, not the tab — the window and the engine are separate, and only the engine touches your files.
Nothing about local-first means no internet, only no upload. The software is delivered as a download and updated the same way, so the machine stays current while the workbook never makes the trip — the network touchpoints are the update, the battery checks, and the model conversation, and none of them uploads the workbook file.
The local architecture also isolates Mica from the rest of your stack. Because nothing is shared with a provider account, your workbook work does not appear in any vendor's activity log, team workspace, or AI history — the record of what you did exists only on your machine, alongside the file itself.
For the full mechanics of how a task plans, authorizes, and runs, how Mica works steps through the flow.
Authorization & Permissions
Connecting Mica to Excel uses a local browser authorization, which lets the tool drive the workbook on your screen. That grant is scoped to Excel on your machine — it does not reach cloud drives, does not send data anywhere, and every task still goes through a plan-approval step before anything changes.
The local-first design also removes a class of problems cloud tools bring to users who cannot rely on cloud services — no subscription to lose, no upload of the workbook, and no server region to worry about.
The grant is reversible. You can disconnect Excel from Mica the same way you revoke any browser app authorization, after which the tool can no longer drive the workbook — combined with the per-task plan approval, control sits at two layers rather than one.
Scoping cuts both ways. The authorization lets Mica drive the workbook, but it does not grant blanket access to the rest of your system: it cannot read unrelated files, browse other applications, or reach cloud storage on your behalf. The grant is deliberately narrow, and the plan approval keeps even the workbook access supervised.
How to Verify Local Processing
- Run a cleaning or processing task and watch the changes happen in your local Excel in real time — cell reads, formula evaluation, and writes are all handled by your device's local process, with no waiting on a cloud upload or download round-trip.
- Watch the plan step: Mica shows the operations it will run before any cell changes, so nothing in your workbook is touched without your explicit approval.
- Revoke the Excel authorization at any time and Mica immediately loses the ability to drive your workbook — the tool's permission stays inside your local client, with no remote channel that keeps working after you withdraw it.
- Read the privacy policy and the compliance statement, and compare the zero-upload commitment they make against what you observe; the two should match.
Verification does not have to be a technical exercise. The transparent plan review and the instant local response let you feel, task after task, that your data is being processed right in front of you. Together they mean security and privacy are not marketing words, but a certainty you hold in everyday use.
Company & Compliance
Mica is developed by MINDLINK TEC LTD, a company registered in Ipswich, UK. Local-first processing is the company’s stated product direction, and this page is where that commitment is documented in public.
Local processing also aligns with the core data-minimization and privacy-first principles behind modern privacy regulations: the workbook that never leaves your device is, by construction, data you never have to transfer, store, or justify. Whatever compliance framework your organization sits under, this design removes cross-border data transfer, cloud-storage registration, and third-party audit exposure at the source.
That posture also simplifies the practical side of compliance: no data-transfer agreements are triggered by a file that never moves, and no retention policy is needed for data that is not stored. The workload is audit-friendly precisely because the processing footprint is a single machine you own.
Software supply chain is the one remaining surface. The installer is signed and delivered through the normal update channel, and because everything runs locally you control when an update applies rather than having behavior change on a server schedule — no third party is interposed between you and the version you run.
On-device processing also inherits whatever protection you already give the machine — a Windows account password, BitLocker on the drive, your corporate endpoint controls. Because the workbook file does not move, the security of the file is the security of the device you already manage, rather than a remote server you do not.