HUD for Claude — Privacy Policy

Last updated: October 19, 2026

HUD for Claude is a desktop app for macOS and Windows, with an optional Chrome extension. Both are built by a single independent developer. This policy explains exactly what is read, where it goes, and what is stored.

Short version

No data about you is sent to the developer or to any third party. There are no analytics, no accounts, no tracking. The app reads files already on your computer, and makes two kinds of network request: asking Anthropic — and, if you use Codex, OpenAI — about your own account, using your own credential that Claude Code or the Codex CLI already stored on this machine; and asking the developer's download server, every few hours, whether a newer version of the app has been published — a request that carries nothing about you.

What the desktop app reads

• Your Claude Code credential, from wherever Claude Code keeps it. On macOS that is the keychain entry Claude Code-credentials, and macOS will ask your permission the first time. On Windows — and on a Mac whose keychain has no entry, or where you declined — it is Claude Code's own file, ~/.claude/.credentials.json (%USERPROFILE%\.claude\ on Windows, or the folder Claude Code has been set to use instead). The app uses it only to call Anthropic's own API on your behalf. It is never copied to a log, to another file, or anywhere off this machine; the one write is putting a renewed credential back where it came from, below.
• Your account name, email, and plan tier — returned by Anthropic when the app asks who the credential belongs to, so the panel can show whose usage it is displaying. Shown on your screen only.
• Your plan's usage allowance — session and weekly percentages, and when they reset.
• Your Codex sign-in, from ~/.codex/auth.json — only if that file exists, which means you have signed in to the Codex CLI with a ChatGPT account. It is used only to ask OpenAI for your own Codex rate-limit windows. If the file is absent, the Codex ring simply does not appear.
• Your local Claude Code transcripts (~/.claude/projects/). HUD reads these files line by line and keeps only each API call's timestamp, model name, and token counts, and for the Activity heatmap, when each message was written and how many tool calls a reply made (counted only; what a tool did is not kept). The text of your prompts and Claude's replies passes through memory while a line is parsed, and is never stored, shown, or transmitted.
• System readings — only if you turn on System Monitor (right-click → System Monitor; it is off until you do). CPU, memory, GPU and disk space, and on a Mac the chip, SSD and battery temperatures. To say what is behind those numbers, it also reads the list of running programs — each one's name, CPU and memory use, disk activity and GPU time — and, for Claude Code sessions and language runtimes such as node or python, their command line and working folder, so a dev server can be named after its project. Read every few seconds while it is on, shown on your screen, and kept only in memory: never written to a file, never sent anywhere.
• Folder sizes — only when you click "Find what's using space" (macOS). The app measures how much room some folders take up: build output and dependencies in your code folders (such as node_modules), caches, Trash, Downloads, Xcode, Docker and the like. It reads their sizes only, never the contents of the files in them, and never deletes anything.

What the desktop app writes

• Back to wherever the Claude credential came from: Anthropic issues a new credential each time the old one is renewed and invalidates the previous one, so the app must store the replacement or Claude Code would be signed out. It writes back only that renewed credential, to the store it read from — the keychain entry, or .credentials.json — and reads it back to verify the write. In the file it replaces only Claude's sign-in section, keeps every other key, and swaps it in atomically, so a half-written file can never sign Claude Code out.
• Back to ~/.codex/auth.json: your Codex sign-in expires about ten days after it is issued, and only the Codex CLI renews it — which never happens on a machine where that CLI is not being run. So when it is within an hour of expiring, the app renews it the same way Codex would and writes the replacement back. OpenAI issues a new credential each time and invalidates the previous one, so the file is rewritten whole, atomically, with every other key in it preserved, and read back to confirm the write before the new credential is used.
• A local cache, usage-cache.json in the app's own folder (~/Library/Application Support/claude-hud/ on macOS, %APPDATA%\claude-hud\ on Windows) — token counters per day and model, message and tool-call counts per day, and when each session started, so the app doesn't rescan every file each time. It holds numbers and times only: no prompts, no replies, no credentials.
• A preferences file — your "Launch at Login" choice, and your pet, notch and update settings.
• Updates, when there is one: the new version is downloaded to a temporary folder and checked against the developer's signature; when you choose Restart to Update, it takes the old app's place and the download is deleted. A few lines recording each update's version and outcome are kept in update.log, next to the preferences file.

Network requests

The app talks to these hosts and no others:

• api.anthropic.com — to read your own Claude usage allowance and to renew your own Claude credential.
• chatgpt.com — to read your own Codex rate-limit windows. Only if you have a Codex sign-in on this machine.
• auth.openai.com — to renew that Codex sign-in. Only if you have one, and only when it is about to expire.
• pub-73af813618eb49f9a18951f929fab6e6.r2.dev — the developer's download server, a Cloudflare R2 bucket — every few hours, to see whether a newer version has been published, and when one has, to download it. The request carries nothing about you or your usage: only what any download carries, your IP address and the app's version in its user agent. Cloudflare runs the server; the developer's dashboard shows how many requests it received, not who made them. To stop it: right-click → Updates → uncheck Check Automatically.

Nothing about you is sent to the developer: the update check asks for a file and sends nothing back. There are no analytics SDKs, crash reporters, ad networks, or third-party APIs of any kind.

The Chrome extension

On macOS it is not needed — the keychain route above supplies the allowance, so there is no browser tab and no claude.ai session involved.

It is no longer needed on Windows either. Claude Code keeps its credential in a file there rather than a keychain, and the app reads that file directly — so the allowance arrives the same way it does on macOS, with no browser tab involved. The extension remains only as a fallback for when the credential cannot be read at all.

The extension reads no other page and none of your conversations. Its cache lives in chrome.storage.local and never leaves your browser profile.

What is never kept, read, or sent

• The text of your conversations and prompts is never kept or sent. HUD has to parse each transcript line to count it; the text is discarded on the spot.
• Your payment method, card, invoices, or billing address are never read.
• No website other than claude.ai's usage page is read (extension only, when in use).

Children

HUD for Claude is not directed at children under 13 and does not knowingly collect any data from them.

Changes to this policy

If the product ever starts reading different data, this policy will be updated and the "Last updated" date at the top of this page will change.

Contact

Questions? Email metasui1491@gmail.com.

HUD for Claude is an independent product and is not affiliated with, endorsed by, or sponsored by Anthropic.