Mushroom EN · 中文 Request early access

Draft: to be reviewed by a lawyer before it is published. Not legal advice.

Privacy policy

Last updated (effective date): [date]

This policy is written in English and Chinese. The English version governs; the Chinese version is for reference only.

1. Who we are and what this policy covers

The sharing service, Workspace AI and usage data of Mushroom are provided by Jiteng Xu, an individual developer based in Shenzhen, China (“we”, “us”). For anything about privacy, write to privacy@pickmushroom.com.

This policy covers: the “Share link” feature in Mushroom and the share pages on pickmushroom.com; project forwarding addresses (you forward a client’s email to a project’s address, and it becomes client feedback in Mushroom); Workspace AI (the AI models we provide in Mushroom for automatic organizing, the AI bar and the agent in Mushroom), whose content passes through our service to the model providers (section 14); the usage data Mushroom sends us (section 12); and the problem reports you choose to send us (section 16).

Early access: if you ask for early access on our homepage, we keep the email address you give, and which AI tools you use if you say, only to email you when your invite is ready; we keep them for 12 months at most and then delete them, and you can ask privacy@pickmushroom.com to delete them sooner.

Mushroom itself runs on your computer and does not upload your library to us. Only these leave your computer: the version you choose to share; the email you forward to a project address; what is sent to a model when you use Workspace AI (for background organizing, a file’s title and first paragraph; when you use it yourself, what you ask it to work on), passed through our service to the model provider and not kept by us; usage data; and the reports you choose to send. Once you connect a mailbox, Mushroom reads it directly on your computer, and your mail does not pass through us (section 13).

Where this applies: The first version is offered to users in the United States, and this policy is written under U.S. law. People elsewhere can download and use it, but we do not market, price or provide local services for any other country or region. Before we offer the service in Europe, the United Kingdom, Japan or elsewhere, we will update this policy and tell you in advance.

2. Two kinds of people

UsersClients
WhoPeople who sign in to Mushroom, create share links or use Workspace AIPeople who open a share link to look at it and leave feedback, without signing up
Their relationship with usWe provide the service to you under the terms of service. We handle your account, usage, device identifier, subscription and usage data as this policy describesWhat you leave on a share page is for the user who shared it with you. We keep it and pass it on for that user, handle it on the user’s instructions, and use it for nothing else

The same applies when a client’s email to a user is forwarded to a project address, and when a file a user gives Workspace AI contains personal information about clients or other people: we process it on the user’s behalf, only to provide the features the user asks for. We do not sell it, share it, use it for training, or combine it with other data we hold. The user must have the right to use this information in this way and, where needed, tell their clients (terms of service, section 3).

3. What we collect and why

DataWhoseWhy
Email addressUsersSigning in (a code by email, or the address Google or Apple gives us when you sign in with them; Apple’s relay address if you chose “Hide My Email”), and telling you about new comments (you can turn this off)
Sharer nameUsersSo clients see who shared the page
Signed-in devices (device name, sign-in method, times)UsersSo you can see and sign out other devices; account security
Display languageUsersTo email you in your language
What you share: the files of the version you chose, its title, “What changed”, who it’s forUsersTo show it to clients
How many times a link was opened, and when it was last openedUsers’ linksShown on the card; who opened it is not recorded
A comment’s text, where it is pinned, and its timeClientsPassed on to the user who shared the page; we process it for that user
A client’s follow-ups (replies under a comment) and their timesClientsPassed on to the user who shared the page; shown under that comment to anyone with the link
The sharer’s replies to clients and their times; which comments are “addressed”, and in which versionUsersShown to clients on the share page (the user’s own copy is what counts; ours is only there to show clients)
When “Approve” was clicked or withdrawnClientsPassed on to the user who shared the page
Delete codes and withdraw codes (only their hashes)ClientsSo whoever left a comment or follow-up can delete it, or withdraw an approval
Email forwarded to a project address: body, attachments, subject, the names and addresses in its sender and forwarding chain, and timesForwarded by users; mostly written by clientsPassed to the user’s Mushroom, where it becomes client feedback in that project; attachments go into the project. We process it for the user
Project forwarding addresses (16 random characters, unrelated to the project’s name) and which project each belongs to (an ID in Mushroom, not its name); registered sender addresses, verification codes (only their hashes), and how many emails arrivedUsersTo tell which user’s project an email is for; to accept only mail forwarded from addresses you registered, so others can’t push things into your address
Content sent to Workspace AI when you use it yourself (the AI bar, “Ask an agent”, handing a task to Mushroom, writing a cover note, the check before sending, placing client feedback): the passage or files you chose, what you asked for, and the project’s files, comments and client feedback the agent reads step by stepUsers (may include clients’ personal information)Passed to the model provider to do what you asked, with the result returned to your Mushroom; passed through only, never kept
Content sent to Workspace AI for background organizing (filing into projects, summaries, search re-ranking, judging whether a file is a deliverable): a file’s title and first paragraphUsersAutomatic organizing; passed through only, never kept; for the model provider, see section 5. You can turn it off at any time (then only this computer’s rules are used)
Content sent to Workspace AI to split a client’s email (the points this computer’s rules can’t split, and the file’s outline)Written by users’ clientsTo split an email into separate points of feedback and put each in the right place; passed through only, never kept. You can turn it off in Settings › Account & email › Email
Device identifier: a random token created the first time you use Workspace AI (kept in your computer’s keychain); after public launch, also Apple DeviceCheck’s two bits (whether this computer has had its free credits, and whether it was flagged for abuse)Users’ devicesTo track free credits and prevent abuse; not used for analytics, and kept apart from the random ID in section 12
Workspace AI usage: device or account, purpose, model, token counts (including those read from the cache), estimated cost, credits charged and time; no contentUsersCredits, billing, reconciliation and overall limits
Subscription (once paid plans start): the subscription status, country, and payment and renewal times Paddle gives us; we never see card numbersUsersStarting and renewing Pro; tax records
Usage data (section 12): a random ID, event names and counts, a summary of each Workspace AI task, error signatures, version and system; no contentUsers’ devicesTo know which features are used, what goes wrong, and whether it got fixed
Reports you choose to send (section 16): the parts you tickedUsersTo find and fix problems
IP addressEveryoneOnly used while a request passes, to limit request rates and stop abuse and password guessing; never written to the database, and no access logs. When you use Workspace AI, also used to see which country you are in (Workspace AI is not offered in mainland China in the first version), and not written down either

What we don’t collect: clients’ names, email addresses, phone numbers or companies (except where written in email forwarded to us); precise location, biometrics, health information, bank card numbers or government ID numbers; any tracking cookies, advertising IDs or third-party analytics; anything in your library you didn’t share (other versions, comments, sources, prompts, file paths); and the content sent to Workspace AI itself (passed through, never written down).

Where it comes from: from you directly (signing in, sharing, forwarding, sending reports); generated as you use Mushroom and share pages (open counts, usage, usage data); from your clients (feedback on share pages, email you forward); from Google and Apple (the email address they give us when you sign in with them); and from Paddle (subscription status, country).

4. Cookies, browser storage and Do Not Track

WhatWhoseWhat forHow long
The share page’s password session cookie (HttpOnly, Secure, SameSite=Strict, for that one link only)Clients opening a link with a passwordNot having to enter the password again after entering it correctly12 hours
The browser’s localStorage: delete codes (for comments and follow-ups), withdraw codes, versions seen, when you last lookedClientsSo you can delete your own comments and follow-ups and withdraw an approval on this device, see “Updated to v4”, and tell which replies are new. When you last looked is the time of the newest item you sawUntil you clear it (one click under “Privacy” at the bottom of the share page)

All of these are needed for features you ask for, so there is no consent banner.

Do Not Track and Global Privacy Control (GPC): We do not track you across websites, and we do not let third parties track you on our pages, so a browser’s “Do Not Track” signal does not change what we do. When we receive a GPC signal, we treat it as a request not to sell or share your personal information; we don’t sell or share it anyway.

5. Who gets it

We give personal information only to the people below and to companies that provide services for us (service providers), and only for the purposes stated. Service providers handle it under their agreements with us and may not use it for anything else.

SubprocessorWhat it does for usPersonal information it handlesWhere data is stored and processedTraining and retention
Cloudflare, Inc. (United States)Hosts the share pages, the API, the database and file storage; runs our relay service for Workspace AI; receives usage data and problem reportsAccounts, links, shared files, comment threads, usage, device identifiers, usage data; problem reports (encrypted only)Database and file storage restricted to the United States (Cloudflare’s U.S. data jurisdiction); requests are handled in the Cloudflare data center nearest the visitorKept as we configure it (privacy policy, section 7); the sharing service logs errors only, kept 7 days
Cloudflare, Inc. (Email Routing)Receives email forwarded to project addresses, rejects spoofed mail by the sender domain’s SPF, DKIM and DMARC, and passes the rest to our serviceForwarded email: body, attachments, and the names and addresses written in itReceived on Cloudflare’s global network; stored in the United States once passed to our serviceDeleted as soon as the user’s Mushroom has collected it
Cloudflare, Inc. (Workers AI)Runs Workspace AI’s models for agent edits and Workspace AI tasks in Mushroom, and for background organizing (filing into projects, summaries, search re-ranking, judging whether a file is a deliverable). The models are open models (DeepSeek V4.1 Flash, gpt-oss-120b, and Kimi K2.6 as the agent’s backup), run by Cloudflare itself, not through the APIs of DeepSeek, Moonshot AI or any other model maker; no data goes to those companiesWhat the user asks it to work on; for background organizing, only a file’s title and first paragraphCloudflare’s global network (GPUs in more than 200 cities); not restricted to the United StatesNot retained after processing; not used for training
Fireworks AI (backup, only when needed)Runs the same open model (DeepSeek V4.1 Flash) when Workers AI is over capacity (rate-limited)As Workers AIUnited States [to check]Not retained and not used for training by default [to check]
ResendSends sign-in codes, new-comment notifications, and the codes for registering a sender address for a forwarding addressRecipient addresses and the text of these emailsSent through Tokyo; data stored in the United States [to check]Sending records, kept for [to check]
GoogleVerifies who a user is when they choose “Sign in with Google” (users only, never clients)The user’s Google account email addressUnited States—
AppleVerifies who a user is when they choose “Sign in with Apple”; relays notification emails when they chose “Hide My Email”; after public launch, DeviceCheck records whether this computer has had its free credits and whether it was flagged for abuse (users only, never clients)The email address or Apple’s relay address; two bits (no content, and they don’t say who you are)United States—
Paddle (once paid plans start; contracting entity [to check])Collects subscription payments and handles sales tax, invoices and refunds; it is the merchant of recordPayment details (it collects them; we never see them), subscription status, country[to check]Under Paddle’s privacy policy and tax law

6. Where data is processed and who can see it

DataWhere it is storedWho can see it
Accounts, links, shared files, comment threads, usage, device identifiers, usage dataCloudflare’s database and file storage, restricted to the United States (Cloudflare’s U.S. data jurisdiction)Us (in Shenzhen, China, through Cloudflare’s dashboard and our own admin page, only to run the service and investigate problems)
Reports you choose to sendAs above, stored encrypted with the developer’s public keyOnly the developer, who decrypts them on the developer’s own computer with the private key (section 16)
Email forwarded to a project addressReceived on Cloudflare’s global network and stored in file storage in the United States; deleted as soon as your Mushroom has collected itOur service passes it on automatically; we don’t open it
Content sent to Workspace AIOnly passes through our relay service on Cloudflare; for where the models run, see section 5The model providers, as section 5 describes; not us (we don’t keep it)
The library on your computer, mailboxes you connect, what is read on your computerYour computerOnly you

Stored in the United States does not mean processed only in the United States: your request first reaches the Cloudflare data center nearest you, where it is decrypted and handled by our service, which then reads and writes the database and files stored in the United States. For where content sent to Workspace AI is processed, see section 5.

We operate this service from Shenzhen, China. When we handle your account, answer your requests, or look at usage data and the reports you send, we access the service providers above from China. China’s laws differ from those of the United States.

7. How long we keep it

DataHow long
Shared filesDeleted at once when the link is revoked; 30 days after it expires (within those 30 days it can be extended and opened again). When the link moves to a new version: with “Client can see earlier versions” on, the latest 3 versions are kept, each deleted 90 days after it was replaced, and any the user takes down are deleted at once; with it off, earlier versions are deleted at once
Comment threads: clients’ comments and follow-ups, the user’s replies, “addressed”Deleted 90 days after the last activity (a new follow-up, a reply, marking it addressed), or with the link 30 days after it is revoked or expires, whichever comes first. A reply the user withdraws, or a follow-up a client deletes, is deleted at once; a comment its client deletes is deleted at once with its whole thread, the user’s replies included
Clients’ approvalsKept on the server for 30 days after the user’s Mushroom has collected them, then deleted; any not collected are deleted with the link
The link itself (title, who it’s for, times)Deleted 30 days after it is revoked or expires
Sign-in codesDeleted once used, or after 10 minutes
Sign-in sessionsEnd when you sign out, or after 90 days unused
AccountsDeleting an account deletes all its links, files and comments at once; the account record itself is deleted 30 days later (so a mistaken deletion can be undone)
Email forwarded to a project address (body, attachments)Deleted from the server as soon as your Mushroom has collected it; if never collected, 30 days after it arrived; at once when you delete your account
Project forwarding addressesDeleted 30 days after being turned off or replaced (mail forwarded in that time is still bounced); at once when you delete your account
Registered sender addressesDeleted when you remove them or delete your account; if registered but never verified, one day after the code expires
Content sent to Workspace AIWe don’t keep it: it is only in memory while being passed on. For how long the model providers keep it, see section 5
Workspace AI usage records[e.g. 13 months], for billing and reconciliation; an account’s part is deleted at once when the account is deleted
Device identifiersDeleted [e.g. 12 months] after last use; Apple’s two bits are stored by Apple, and we clear them after [to be decided]
Usage data (daily counts, task summaries, error signatures)Deleted after 180 days
Reports you choose to sendDeleted after 90 days, or at once when you click “Delete”; the copy the developer downloads to their own computer to read is deleted once the problem is dealt with
AI request logs synced automatically during the trial (only if you turned it on)Deleted after 30 days; syncing stops when the trial ends, and the rest is deleted within 30 days
Subscription recordsKept as tax law requires [to check: how many years], by Paddle and by us
LogsThe sharing service logs errors only (no email addresses, no content), kept 7 days in Cloudflare, not one line per request; we keep no access logs of our own; Workspace AI’s relay service writes no logs. How long Cloudflare keeps its own request metadata (URLs, times, firewall events): [to check]

The comments a user’s Mushroom has collected are kept on the user’s own computer, not with us; whether to delete them is up to the user.

8. Your rights

Whichever U.S. state you live in, we give you the rights below (the rights that California and other states with privacy laws require, we give to every user alike):

RightWhat it means
To know and accessWhich personal information we have about you, where it came from, why we have it and who we gave it to; and a copy you can take elsewhere
To deleteDeleting the personal information we hold about you (except what the law requires us to keep, such as tax records for a subscription)
To correctFixing what is wrong
To opt out of sale and sharingWe don’t sell or share it, so you don’t need to ask
To limit the use of sensitive personal informationWe use it (for example, the content of other people’s email forwarded to us) only to provide the features you ask for, never to infer characteristics about you, so there is no separate switch
Not to be treated differently for using these rightsYou won’t get fewer features or pay more because of it
To appealIf we turn down your request, you can reply to appeal; we will answer within 45 days and explain why. If you are not satisfied, you can contact your state’s attorney general

9. Security

Link addresses are 128-bit random and can’t be guessed; passwords are stored only as hashes; shared web pages run on a separate domain in a sandbox and can’t read anything on the share page; every connection uses HTTPS; session tokens and mailbox authorizations are kept in the keychain of the user’s computer; reports you choose to send are encrypted with a public key, and the server holds only the encrypted copy. No method is perfectly secure. If there is a data breach, we will notify the people affected promptly as the applicable state laws require, and the state attorney general or other authorities where required.

10. Children

The service is meant for working adults and is not directed to children under 13, and we do not knowingly collect their personal information (the U.S. Children’s Online Privacy Protection Act, COPPA). If you believe we have collected information from a child under 13, write to privacy@pickmushroom.com and we will delete it. The terms of service require users to be at least 18 (terms of service, section 1).

11. Changes

If this policy changes in a way that matters, we will tell you in Mushroom and here at least [30] days in advance, and change the “Last updated” date above. If we want to use data we already have for a new purpose (for example, if we once said “not used for training” and later want to train on it), the change only applies to data collected after it, unless you first give us your express consent. We will never quietly change the terms to reuse old data.

12. Usage data from Mushroom (no content)

By default, Mushroom sends us usage data, so we know which features are used, what goes wrong, at which step Workspace AI tends to fail, and whether it got fixed. It contains no content, and you can turn it off at any time.

What is sentWhat is not
Event names (such as “new arrival”, “handed to agent · feedback file written”, “sent · shared”, “saved”) and counts, totaled by dayFile contents, file names, paths, titles, project names, comment text, prompts, the model’s answers
One summary line for each Workspace AI task, in fixed words and banded numbers only: where it started, which way it went (ours, an agent borrowed on your computer, your own key), which kinds of files, how many steps, which tools were used and whether they worked, how it ended, the category of any failure, at which step and tool it failed, how many times it asked you, banded time and token counts, whether you kept the result or discarded it, the reason you clicked when you discarded it, and how many changes you undid after keeping it; also a random 6-character code, shown on that line when a task fails, so if you tell us the code we can find the lineWhat you asked for, what the AI said, the tools’ arguments, what was read or changed, model names, error text
Error signatures: the area where an error happened, its type, and the place in our own code, sent at most once a day with a countError messages (which often contain paths and file names)
Mushroom’s version, the macOS major version, the chip, the display languageComputer name, user name
A random ID (created the first time Mushroom opens, and replaced when you clear the data), to keep one computer’s data togetherAccount, email address; this ID is not linked to your account or email address, nor to Workspace AI’s device identifier; the IDs of cards, projects and share links
The date (which day; no hours or minutes)The exact time

13. What Mushroom reads on your computer (not uploaded)

The first time it opens, Mushroom starts looking on your computer for files made by AI, without first asking you to choose where or to agree; when it’s done, it says at the top how many it found, what it read and where to turn it off, and you can undo what it found in one click. All of this is read on your computer and is not sent to us. Once files are in Mushroom, background organizing sends their titles and first paragraphs to Workspace AI to file them into projects and summarize them (section 14; on by default everywhere, and you can turn it off); content leaves your computer otherwise only when you share it or ask an agent to change it (section 3, and the explanation in Settings › AI & models).

What it readsWhyHow muchHow to turn it off
Agents’ local records: Claude Code (~/.claude/projects), Codex (~/.codex), WorkBuddy (~/.workbuddy-ai)To recognize which files they wrote, and from which of your requests; to see what they are doing right nowWhich file was written, when, and from which request; for their status, only fields such as tool names and times. The request you made is kept in your local library and shown on the cardSettings › Collect & organize, “Read Claude Code, Codex, and WorkBuddy session history”
The “Where from” of files in Downloads (the download address macOS records)To recognize which AI website a file was downloaded fromOnly the address, to tell which service it was; files downloaded from anywhere else are not taken inRemove “Downloads” in Settings › Collect & organize
Agents’ own folders (~/Documents/Codex, ~/Claude, ~/WorkBuddy AI and so on, read only if they exist)What is in them was made by that agentThe files themselvesRemove them in Settings › Collect & organize
Files taken inTo show, compare and keep versionsThe card points to the original file; files from Downloads are copied into your library (on your computer)“Delete” on the card, then “Remove from Intelligent Workspace”
Mailboxes you connect (in Settings › Account & email › Email: IMAP with an app password; Gmail or Outlook / Microsoft 365 with one-click sign-in, read-only scopes gmail.readonly and Mail.Read; or Apple Mail)To put clients’ replies into the right project as client feedback, with attachments as new versionsOnly the body and attachments of two kinds of mail: mail from clients you have sent to, and replies to mail you sent. In Sent, only mail from the last 30 days with a project file or share link, and only to note the recipients (to recognize clients). Other mail is filtered out on the mail service’s side by sender and conversation (for Apple Mail, inside Mail), and not downloaded. Nothing is marked read, moved, deleted or sent“Disconnect” in Settings › Account & email › Email, then delete the app password in your mailbox’s settings (for one-click sign-in: disconnecting Gmail also tells Google to revoke access; for Outlook, revoke it under “Apps and services” in your Microsoft account. For Apple Mail, turn it off in System Settings › Privacy & Security › Automation)

14. Workspace AI (the models we provide)

Automatic organizing, the AI bar and the agent in Mushroom use the models we provide by default (called “Workspace AI” in the interface). You can also borrow an agent you are signed in to on your computer (Claude Code, Codex), or add your own API key, which is then used first.

CaseWhat is sentDefaultWhat it says beside the result or in the explanationHow to turn it off
You use Workspace AI yourself (“Ask an agent”, Workspace AI tasks, handing a task to Mushroom…)The passage or files you chose and what you asked for; the files, comments and client feedback in this project the agent reads step by stepSent when you click“Workspace AI · not kept, not used for training”Turn off “Use Workspace AI” in Settings › AI & models; excluded projects and locations are never sent
Background organizing (filing into projects, summaries, search re-ranking, judging whether a file is a deliverable)A file’s title and first paragraphOn (everywhere): explained the first time Mushroom opens (on the home page’s “Found N files made by AI” card, or on the first page when nothing was found)“Not used for training; the model’s maker may keep it a while against abuse (usually up to 30 days)”, with the details in this policy“Use Workspace AI for organizing in the background too” in Settings › AI & models; exclude projects and locations
Splitting a client’s email (end of section 5)The points this computer’s rules can’t split, and the file’s outlineOn (everywhere)A line of explanation in Settings › Account & email › EmailTurn it off in Settings › Account & email › Email; the whole email then goes on the whole file
Borrowing an agent on your computerAs when you use it yourselfBorrowed when you use it yourself; borrowing it for background organizing is off by default, explained once when you turn it on“Your Claude Code · model name”, with “Data is handled under your settings with Anthropic” beside it (OpenAI for Codex). We don’t say “not used for training” for it: it is your own account, under your settings thereSettings › AI & models
Your own API keyAs when you use it yourselfUsed first once added“Your key · model name”, under that provider’s API termsDelete the key

15. Google user data

16. Reports you choose to send us

17. Residents of other states

The privacy laws of Virginia, Colorado, Connecticut, Utah, Texas, Oregon, Montana, Iowa, Delaware, Nebraska, New Hampshire, New Jersey, Tennessee, Minnesota, Maryland, Indiana, Kentucky, Rhode Island and other states give residents the right to access, delete, correct and take a copy of their personal information, to opt out of targeted advertising and sale, and to appeal. Whether or not these laws apply to us, we give users in every state the rights in section 8, including the right to appeal. We do not sell any personal information and do not do targeted advertising.

Nevada residents can ask us not to sell their personal information by writing to privacy@pickmushroom.com; we don’t sell it anyway.