For IT
Contacted only when someone uses it
- The AI provider chosen in Settings, and only that one:
generativelanguage.googleapis.com(Gemini),api.anthropic.com,api.openai.com, your<resource>.openai.azure.com, orbedrock-runtime.<region>.amazonaws.com. A local model (Ollama) stays on the computer. - Code hosts, for pull requests and sharing branches:
github.comandapi.github.com, or your own GitHub Enterprise or GitLab address. - Package registries while building or testing a project (
registry.npmjs.org,pypi.org). A package the AI names that the project does not already list is installed only after a person allows it. - Deploy and email services only when an account for them is added: Render, Vercel, Resend, or your mail server.
Contacted on its own, and how to turn it off
- Update check:
api.github.com/repos/shrestha-droid/branchhq, the version number only. Off with the environment variableBRANCH_HQ_UPDATE_CHECK=off. - Timestamping the audit record:
timestamp.digicert.comreceives a fingerprint (SHA-256) of the record, never its content. Off withBRANCH_HQ_TIMESTAMPS=off; the record is still kept and hash-chained, without the outside timestamp. - The chosen AI provider's list of models, at most once a day, to keep model names current.
Network
- Outside calls use the computer's own proxy settings (including PAC files and proxy sign-in) and the operating system's trusted certificates, so a company proxy and TLS inspection work as they do for the browser.
- No port is opened and nothing is announced on the local network unless someone uses device pairing (the Devices panel): then TCP port 47821 (or a free port) and mDNS/Bonjour. Every paired message is signed, and work sent from another device runs only after a person approves it.
- No analytics, tracking or crash reporting is sent anywhere.
Data
- Projects, chats, settings and the audit record stay on the computer, in the app's data folder (
~/Library/Application Support/branch-hqon macOS,%APPDATA%\branch-hqon Windows). - API keys and code-host tokens are encrypted with the operating system's keychain (Keychain on macOS, DPAPI on Windows).
- The privacy record lists every request sent to an AI provider: when, to which model, how much, and its fingerprint. The optional privacy shield replaces personal details and secrets with placeholders before anything is sent, and puts them back in the answer.
- Commands the AI runs on a repository run in a throwaway copy, one plain command at a time with no shell, with secrets removed from their environment. Pushing, accounts, privileged and network-shell commands are refused.
Need this in your own security questionnaire's format? See the licence, or ask us.