Privacy
Bower runs on your machine. There is no Bower account, no Bower server holding your documents, and nothing of yours stored anywhere we control. This page describes every case where something does leave your computer, including the ones that are easy to overlook.
The short version
Your files never come to us. They are opened, edited and saved on your own disk. We operate no servers that could hold them.
When you ask the assistant to do something, that part of your document goes to the AI provider you chose — under your own account, on their terms. Not to us. If you run a local model, it goes nowhere at all.
We count how the app is used, never what you write. Anonymous feature counts, described in full below, and switchable off in Settings.
What stays on your machine
- Your documents. Spreadsheets, documents, slides and PDFs are read from and written to your disk. Bower has no upload path.
- Your API keys. Stored in Bower's settings file in your user data directory. They are sent to the AI provider they belong to, and to nobody else — never to us.
- Your chat history with the assistant, your custom rules, and your skills.
What leaves your machine, and where it goes
Five things, and only these.
1. What you ask the assistant
When you send the assistant a request, it sends your instruction and the relevant part of the document to the AI provider you configured — Anthropic, OpenAI, Google, DeepSeek, Mistral, xAI, Groq, Moonshot, OpenRouter, any OpenAI-compatible endpoint you name, or your existing Claude subscription through the Claude Code CLI. This is the same exposure as pasting that text into that provider's chat window, and it is governed by your agreement with them, not with us.
If you point Bower at a model running locally on your own machine, nothing leaves at all.
2. Web searches the assistant makes
When the assistant searches the web or looks up an image, the search query goes to a search provider — Serper, a commercial Google Search API, or DuckDuckGo when Serper is unavailable. The assistant writes those queries from your request, so a query can contain words taken from what you asked or from your document.
When the assistant reads a web page you or it selected, Bower fetches that URL directly, and that site sees the request as any browser visit would.
Search happens only when the assistant decides a request needs it. It is never automatic and never in the background.
3. Connectors
If you connect an external service, whatever the assistant sends to that service goes to that service. Bower does not proxy it and does not keep a copy.
Two things are worth knowing: Bower reads and writes the connector list belonging to your Claude Code CLI rather than keeping its own, so a connector you add in Bower is added for your whole machine, not just for Bower. Their access tokens live in that CLI's own credential store — Bower never holds them.
4. Update checks
Bower asks our update server whether a newer version exists. That request necessarily reveals your IP address, your current version and your operating system. It contains nothing else, and no document ever travels with it.
5. Anonymous usage counts
Described in full in the next section, and switchable off.
Usage analytics, in full
We need to know whether people use Bower after installing it, and which parts. Without that we are guessing about what to build. So Bower counts events — never content.
Every event carries a random identifier generated on your machine at first run. It is not derived from your hardware, your name, your email or anything else about you, it is not linked to your signup, and clearing it in Settings issues a new one. Events are batched and sent once a day to PostHog, our analytics processor, on their European infrastructure. Your IP address is used once, on arrival, to work out roughly where the request came from — country, region and city — and is then discarded. We keep that rough location, because it decides where we put launch effort and support hours. We do not keep the IP address, and we do not keep your postcode, your coordinates, or anything else that would place you more precisely than the city you are in.
| What we record | Example |
|---|---|
| App version, operating system and processor type, language | 0.5.5, macOS, arm64, en |
| That the app was launched, whether it was the first launch, how long the session ran, and how much of that you were actually using it | first run, 42 min open, 12 min active |
| Which app you opened | sheets |
| Which kind of file was opened, and a rough size band | xlsx, medium |
| Which AI provider you connected | anthropic |
| That you sent the assistant a message, how many tools it used, and how long it took | 3 tools, 12s |
| Whether you accepted or rejected its edit | accepted, 7 changes |
| That a skill or connector ran, and how often — by name only when it is one of ours | built-in skill, 2× |
| Which onboarding steps you completed or skipped | connect, completed |
| Whether an update check succeeded or failed, and a code if it failed | installed · failed, E_NETWORK |
| Roughly where you are, worked out from your IP address and kept as country, region and city — never finer | Bengaluru, KA, IN |
| Errors and crashes, as a code | E_PARSE_FAILED |
Alongside the rows above, each batch carries the name and version of the analytics library that sent it, and a flag marking it as sent by the app rather than a web page. Those describe our software, not you.
What is never recorded
- The contents of any document, cell, slide or page
- File names, folder names or file paths
- Anything you type to the assistant, and anything it replies
- API keys, or the address of a custom or local endpoint
- The names of skills you wrote yourself, or of connectors you added
- Error messages — only the code, because messages tend to contain file paths
- Your name, your email, your IP address, your postcode, or your coordinates
This is enforced in the code, not by convention: every event passes through one function that drops any field not on the list above, and a test fails the build if that list and this page disagree.
Turning it off
Settings → Privacy → Usage analytics. Switch it off and Bower stops sending, immediately and permanently, with no other change to how it works. Nothing is queued for later. You are told this on first run, before you have used the app.
If you gave us your email
The download form on this site records your email address, whether you asked about the Windows build, which page you came from, and the time. It is kept in a private spreadsheet, used to send you the download link, release notes and word of a launch — and never sold, rented or shared.
Every message has a way out, and replying to any of them works too. Ask us to delete your address and we will delete the row.
Children
Bower is a tool for professional work and is not directed at children under 13. We do not knowingly collect anything from them.
Changes
When Bower starts sending something new, this page changes in the same release — that is a rule we hold ourselves to, not a courtesy. Material changes are called out in the release notes rather than left here to be discovered.
Contact
Questions, deletion requests, or anything on this page that does not match what you observe: hello@bower.page. If you find a discrepancy between this page and what the software does, we want to hear it — treat it as a bug report.