Short, because there's little to say.
The app
Retriever has no telemetry, no crash reporter that phones home and no account. Apart from the HTTP requests you send, it talks to one thing: once a day it reads a small file, fetchretriever.app/appcast.xml, to see whether a new version is out, and downloads the update from the same site if you install one. That request carries the app's version in its user agent and nothing else: no identifier, no system profile, nothing about your requests. Turn it off in Settings → General → Updates. There is no license to check and nothing to activate. The one exception is a report you choose to send: only when you choose Report an Issue… (Help menu or Settings → General), Retriever opens a report form hosted by SnagFrog, our issue tracker, and passes along the app version, your macOS version, Mac model, processor type and language. If macOS wrote a crash report for Retriever in the last 7 days, that file is attached too: it holds stack traces, loaded libraries and file paths on your Mac, with your home folder written as ~. Before anything is uploaded, a window lists these details and the file, and lets you read the file and leave it out; nothing leaves your Mac until you choose Continue in Browser. After a crash, Retriever asks once on the next launch whether you want to send such a report; nothing is sent unless you choose to. None of your collections, requests, responses, environments or tokens are included, and no screenshot of the window is taken. What you type into the form, and any screenshots you attach, are used only to answer you and fix the problem. Reports are stored in the EU (Frankfurt), and replies are emailed through Resend in the US; see SnagFrog's privacy policy. If you add a key for the optional AI features, it talks to that one provider, and only then. Your collections, environments, response history and cookie jar are plain JSON files in ~/Library/Application Support/Retriever/ on your Mac. Cookies are only ever sent back to the sites that set them.
The bundled MCP server is a local program. It talks to your AI agent over stdio, reads and writes the same workspace files, and sends the API requests your agent asks for. Nothing else. The AI client you connect may send tool results to its model provider under its own privacy settings. Variable values whose names look like secrets are masked in its answers unless the agent explicitly asks to reveal them.
Optional AI features
Retriever can explain failed responses, suggest chain references and search ⌘K by description. These features are off until you add your own API key in Settings → AI, and each has its own switch. You choose which provider answers, and Retriever talks to that one only: TypeSafe (api.typesafe.ai), Anthropic (api.anthropic.com) or OpenAI (api.openai.com). When a feature runs, it sends that provider the request and response involved (or, for search, your search text and the names, methods and URLs of your saved requests) and the names of your environments. Before sending, it replaces Authorization and cookie headers, fields named like token, secret, password or key, and values that look like credentials with «masked», and cuts bodies to a few thousand characters. Each question asks for one typed judgment rather than prose. The provider's own privacy policy covers what they do with it, including whether they retain it. Your key is stored on your Mac in typesafe.json, readable only by you. Without a key, none of this is sent.
This website
This site uses Google Analytics to count visits and see which pages and buttons are used, but only after you say yes on the cookie banner. Until then, and if you say no, Google Analytics is not loaded at all: no script, no cookie, no request to Google. If you say yes, it sets its own first-party cookies so a returning visit counts as one visitor and not two; the events carry the page, the text of the button and how far you scrolled, never a name, and advertising signals are switched off. Your answer is kept in your browser's local storage, not in a cookie, so saying no does not itself set one; a browser that sends Global Privacy Control is taken as a no without asking. You can change your choice at any time, also from the Cookies link in the footer; going from yes to no removes the analytics cookies there and then. See Google's privacy policy. There are no other trackers. The “Report a bug” button opens the same SnagFrog form; it includes this page's address, your browser and any recent page errors. It is a static site hosted on Netlify, and like any web host Netlify keeps standard server logs (IP address, time, requested URL, user agent) for a limited period. See Netlify's privacy policy. The site uses system fonts, with no external font requests.
If you email me, I use your address to reply. I don't share it, and you can ask me to delete it at any time: gorgekara@gmail.com.
Last updated 18 October 2026 (Retriever no longer takes payments of any kind).