Privacy policy
Last updated 2026-10-08
What this covers
agentleFS is a place for a team to keep what it knows, so that people and the AI agents working for them can find it — with the same permissions applying to both.
This policy explains what we collect, why, and what we will never do with it. It applies to the agentleFS service at agentlefs.com and to the connectors and integrations that are part of it.
Who we are
agentleFS is operated by AGENTLEFS, INC., a Delaware corporation (United States), which is the controller of the personal data described in this policy. Write to us at privacy@agentlefs.com.
What we collect
Mostly two kinds of thing: the account details needed to know who you are, and the content you choose to put in or connect. On our public pages we also count visits.
- Meta harness waitlist — the email address you submit and your request for an early-access notification. Clerk stores the address; we use it to contact you about meta harness early access. Ask us to remove it at privacy@agentlefs.com.
- Account information — your name and email address, from whoever you sign in with. We use it to identify you, to show who wrote what, and to contact you about the service.
- Content you create or upload — the documents in your organization, their history, and who you have granted access to.
- Content you connect — from Google Drive or GitHub, when you connect them. See the Google section below.
- Operational records — an audit log of what was read and changed, which is a feature of the product rather than a by-product: it is how an organization can answer who saw what.
- Visits to our public pages — the page you landed on, where you came from, and your browser and approximate location, counted by Google Analytics. This is collected only on the home page, docs, support, privacy policy and terms, and the public catalogs of agent instruction files and published packages, never inside the console. See Service providers below.
- Addresses you invite — when you invite someone to agentleFS from the console, their email address, who invited them and when. We use it to send that one invitation, to apply the limit of 20 invitations a day, and to note that their account came from your invitation if they sign up. We delete it 30 days after the invitation. A folder you shared with them along the way is a share like any other, kept until it is accepted or withdrawn.
- How you found us — when you create an account, the name of the site that sent you (never the full address) and any campaign tag on the link you arrived by. It is recorded once, kept in our own database, and not shared with Google Analytics or anyone else.
Google user data
If you connect Google Drive, agentleFS reads the folders you choose so their documents can be searched and read inside your organization. This is optional, and it is scoped: you pick the folders, we read those, and nothing else in your Drive is touched. You can disconnect at any time.
We request one Google scope: drive.readonly. It is read-only. agentleFS never creates, edits, moves or deletes anything in your Google Drive. When you choose to edit a document, we send you to Google's own interface, signed in as yourself — that edit happens in Google, not here.
We request read access to the whole Drive rather than a narrower scope for two reasons. The feature mirrors a folder continuously — after you connect it, other people add files, rename them and edit them, and those changes have to arrive without you picking each file again. And in an organization it mirrors the folder's sharing as well, so that a document restricted in Drive stays restricted here. Google's narrower alternatives cannot express either: drive.file reaches only the files one person selects individually through Google's picker and never the ones a colleague adds afterwards, and drive.metadata.readonly withholds the document contents we index.
- What we read: the files in the folders you connect — their contents, their names, and, for an organization, who Google says may read them.
- Why we read the sharing: in an organization, so a document restricted in Drive stays restricted here. Who may read a mirrored document in an organization IS its Drive sharing — agentleFS re-reads it on every sync, so removing someone in Drive removes them here, and adding someone in Drive adds them here. Nobody can be given access to a mirrored folder from inside agentleFS; the attempt is refused and points you back to Drive.
- Where it goes: our own systems, inside your organization's boundary. It is encrypted in transit. It is not sold, never used for advertising, and never used to assess creditworthiness. Two things do leave, and they are described below: the text sent to Voyage AI so search can find it, which Voyage does not keep or train on, and session recording, which captures names and never document contents.
- Who can see it: in an organization, exactly the people the Drive folder is shared with, plus a manager of the organization — who can already read everything in it, which is why only a manager may connect a Drive folder. In a personal workspace there is nobody else, so nothing is mirrored and only you can see it.
- How long we keep it: a copy stays in agentleFS until you remove it. Disconnecting a connector stops us reading anything further from Drive, but it deliberately leaves the documents already synced in place, so the people who could read them still can. Removing the folder is what takes the content out of use, and Delete forever on it destroys our copy.
- When you delete your account: we revoke our access at Google for every Drive connection you made, so agentleFS disappears from your Google account's list of connected apps. Drive documents synced into an organization's folders belong to that organization and stay there for the people Drive shares them with, but nothing further is read through your connection.
- Revoking our access from Google directly, at myaccount.google.com/permissions, also stops every further read; it does not delete what was already synced, which you remove here.
AI models and your data
We do not use your content to train or improve AI models. Not ours, and not anyone else's. It is never included in a training set.
agentleFS exists so an AI assistant can answer questions using what your organization has written down — but the answering does not happen here. Your assistant connects to agentleFS, and we serve it the documents you have given it access to, using your own credentials. What that assistant does next, and which model provider it uses, is between you and whoever you connected. agentleFS makes no call to any model to answer your questions.
Search is the one exception, and it does not answer anything. To find documents by meaning, we send the text of your documents, and each search you run, to Voyage AI. Voyage turns text into embeddings (numbers that represent meaning) and scores how relevant each result is to a search, then sends the result back. It keeps none of that text after processing it and never uses it to train a model. The embeddings are stored in our database beside the document and are deleted with it.
Nothing else here calls a model. We once sent folder metadata to an outside model, to compose one written sentence about who could reach a folder. That feature has been removed. Nothing replaced it.
Beyond that and the session recording described below, no human at agentleFS reads your content except where you ask us to for support, or where the law requires it.
GitHub
If you connect a GitHub repository, we read the files in it so they can be searched and read inside your organization. We request the narrowest permissions that allows: repository metadata, and read access to contents. We do not write to your repositories.
Signing in with Facebook
You can sign in to agentleFS with your Facebook account. Our sign-in provider, Clerk, handles that sign-in with Meta on our behalf.
- What we receive from Meta: your Facebook user ID for this app, your name and your email address. Clerk keeps the Facebook user ID and the access token Meta issues, to link your Facebook account to your agentleFS account, and may keep the profile picture Facebook supplies. agentleFS itself never reads the access token or the picture, and never stores your Facebook user ID in its own database. The only profile pictures we read are those of our own staff who write on our blog, shown beside their posts.
- What we use it for: signing you in, and setting up your account with your name and email address, which are then used as "What we collect" describes. Nothing else: we do not read your friends, posts, photos or anything else on Facebook, and we never post to it.
- We do not sell it, license it, or use it for advertising, and we share it only with the service providers named below, to run the service.
- How long we keep it: the link to your Facebook account stays at Clerk until you disconnect it or delete your account. Your name and email address stay until your account is deleted, apart from backups, which expire within about ninety days.
- Disconnecting: remove agentleFS in Facebook under Settings & privacy → Settings → Apps and websites. Meta then tells us, and we remove the link to your Facebook account at Clerk. Your agentleFS account, documents and any other way you sign in are kept.
- Deleting your data: delete your account in the console, with Delete account forever (see "Delete your account" below), or send a data deletion request from Facebook's Apps and websites settings. Either starts the same deletion. A request from Facebook gives you a confirmation code, which opens a page showing the request's status.
Service providers
These companies process personal data on our behalf, only to deliver the service and under agreement with us. If we add or change one, we will update this page.
- Clerk — sign-in and storage of meta harness waitlist email addresses. It holds your name, email address and sign-in methods. If you sign in with another account, such as Facebook (Meta), Clerk receives your name, email address and account identifier from that provider and keeps the linked account so you can sign in with it again; removing agentleFS in Facebook's settings unlinks it. It also receives the address of anyone you invite who does not have an account yet, and sends the invitation to join an organization itself.
- Railway — hosting. The service and its database, and so everything described in this policy that we store, run on Railway.
- Voyage AI — search. It receives the text of your documents and of each search, and returns embeddings and relevance scores; it keeps none of the text and never trains on it. "AI models and your data" above says more.
- LogRocket — session recording, on agentlefs.com only. It identifies a signed-in person by name and email address, and records how they move around the interface: what they clicked, what errored, and the names of the folders and files on screen. It also records what you type into search. It does NOT capture the contents of your documents: the editor, version history and the excerpts in search results are redacted before anything is recorded, so document text never reaches it.
- Resend — email. It sends the emails about your account, such as the confirmation when an account is deleted, and the invitations to agentleFS or to a shared folder, and share notices, that you send to other people, so it receives the recipient's address and the message.
- Slack — our internal notices. When someone signs up, our team is told their name and the domain of their email address; when someone asks to be let in to Google Drive, the address they gave and their note.
- Google Analytics — counts visits to our public pages (the home page, docs, support, privacy policy and terms, and the public catalogs of agent instruction files and published packages) so we can see how people find us. It runs only on those pages: it is switched off on the sign-in pages and everywhere inside the console where your folders and documents appear, and is not linked to your account. We turned off Google's advertising features, so it is not used to show you ads.
Services you connect
Google Drive and GitHub are not our service providers: they hold your data on their own terms, and you connect them. When you do, we read from them on your behalf, using the access you granted, as the sections on each above describe. Disconnecting stops that.
What you control
Your content is yours, and the controls are meant to be usable rather than nominal.
- Disconnect a connector at any time, with Remove on the folder it fills. Nothing further is read from it. Removing a Google Drive connector together with its account also revokes our access at Google when you are the person who connected that account and no other organization in agentleFS still uses it; otherwise the console tells you so, and you can revoke it at myaccount.google.com/permissions.
- Choose which folders a connector brings in, and change that selection later.
- Delete content. Deleting a document or folder takes it out of every listing, search and read straight away, and nobody can reach it. It stays recoverable by you, so an accidental delete is not a disaster, which also means the underlying copy is still held.
- Delete content forever. A manager of a document or folder (someone who can share and manage access to it, and edit it) can choose Delete forever in the console. It destroys the content, every earlier version, its comments, its name, and what agents attached to it: the decisions, discussions, claims and lessons about it. Nothing in the service brings it back; the one exception, a restore from backup, is set out under "How long we keep things".
- Delete your account forever, yourself, from the console. See "Delete your account" below for what that removes and what it leaves.
- Delete a shared organization as a whole, by writing to privacy@agentlefs.com. There is no self-service control for that yet, and we would rather say so than point you at a button that does not exist.
- Revoke our access from Google directly, at myaccount.google.com/permissions — you do not have to come here to do it.
- Ask us what we hold about you, or ask us to delete it, at privacy@agentlefs.com.
Delete your account
You can delete your agentleFS account yourself, at any time. In the console, open the menu at the top right, choose Delete account, then Delete account forever, and type "delete my account" to confirm. You are signed out everywhere at once, and every agent, key and connected AI client of yours stops working at once. The deletion itself becomes final seven days later: if you sign back in before then, you are offered Cancel deletion, which restores your account as it was, except that the agents, keys and connections that were ended stay ended. After the seven days it cannot be undone. If you asked through Facebook's "delete my data" request, the same seven days apply, and the confirmation code Facebook shows you opens a page with the request's status. Only you can do it, from your own signed-in session: an AI agent connected to your account cannot. If you cannot sign in, write to privacy@agentlefs.com from the address on the account and we will delete it for you.
What is deleted: your personal workspace and everything in it, meaning every document and folder, every earlier version, every comment, your agents and their keys, the connections you made, anything you shared from it with other organizations, anything it published to the public registry, and its activity log. Your sign-in account is deleted too, which signs you out everywhere and ends every AI client's access. We revoke our access at Google for any Drive connection you made, delete the record of how you found us, and send one email to confirm, after which we do not write to you again.
In an organization you share with others, what you wrote belongs to that organization and stays there, the way a shared drive's files do. What you owned there passes to one of its managers, or to another member if it has no other manager, who then becomes its manager. You leave it: your access, your agents there and their keys end, and your name and email address are removed from its list of people, from your agents' names and from your comments. Messages you sent there stay in its conversations as part of its record, under the name you sent them with. Its activity log keeps its entries for what you did there; they refer to you by an identifier, and an entry written while you were a member can still include your name or address as it was recorded then.
What remains: one row noting that an account existed and was deleted, with no name, no email address and no content, so the deletion itself is on record. Copies in our backups expire on the schedule under "How long we keep things", within about ninety days.
If any of your data is under a legal hold, the deletion is refused before anything is changed, and you are told so. Write to privacy@agentlefs.com and we will tell you what applies and when it ends.
Some records are held by the providers named above. When your account is deleted we automatically ask LogRocket to delete every session recording of you and to stop recording you, and we ask our email provider, Resend, to delete any contact record of you; we repeat each request until the provider confirms it. Resend's log of the emails we already sent you cannot be deleted through its API and expires under Resend's own retention; ask us at privacy@agentlefs.com and we will ask them to remove it sooner. Google Analytics runs only on our public pages and is not linked to your account. A GitHub App you installed stays installed on your GitHub account until you remove it in GitHub's settings; we stop using it when you delete your account.
Requests for your data
We disclose personal data to a government or other third party only when legally required by valid legal process, such as a subpoena or court order. Each request is reviewed for legal validity and scope, and we narrow or challenge requests that are overly broad. Unless legally prohibited, we notify the affected user before disclosing so they can object. We do not give any government or third party direct, bulk or back-door access to our systems or your content. Questions: privacy@agentlefs.com
How long we keep things
We do not delete your content on a schedule, and nothing here expires on its own. There is no timer on a document, no archive tier, and no job that removes old material. What you put in stays until you or your organization takes it out.
The one exception is an agent's drafts: the unpublished working notes an agent keeps while it works. When an agent is retired it gets an hour to publish anything worth keeping, and then its drafts are removed; an agent that stops suddenly loses them at once. Anything it published stays, because what it published is a copy.
Invitations are the other thing on a timer. When you invite someone to agentleFS, the record of their address, who invited them and when is deleted 30 days after the invitation, whether or not they ever signed up.
Deleting a document or folder withdraws it immediately — it leaves every listing, search and read, and nobody can reach it — and it stays recoverable by you. That is the point of the two words: the underlying copy is still held, which is what makes an accidental delete survivable.
Delete forever is the other one. It destroys the content, every earlier version, its comments, its name and what agents attached to it, at once, and nothing in the live service brings it back; backups are the one caveat, below. The only thing that stops it is a legal hold: while one applies to a document or folder, a Delete forever is refused and says so, rather than being queued for later.
One record survives a Delete forever, and we would rather say so here than have you find out from a support reply: a row noting that the object existed, what kind of thing it was, who owned it, and who deleted it and when. It carries neither its name nor its path. Our activity log is tamper-evident, so its entries are not removed: the ones about what was deleted keep their place, with their content, name and path blanked. Destroying the record of a deletion because the deletion happened would defeat the purpose of keeping one, so that row stays.
Backups are the honest caveat on both "immediately" and "nothing brings it back". Our hosting platform takes encrypted backups on a rolling schedule — daily kept for six days, weekly for about four weeks, monthly for about three months. Content you delete forever is gone from the live service straight away, and can still sit in a backup that has not rotated out yet, for up to roughly ninety days. We do not mine backups, and we do not restore one to recover something a person chose to delete forever; they exist to bring the service back after a failure. A restore of that kind puts the service back as it was when the backup was taken, so content deleted forever since then would come back with it.
Deleting your account signs you out at once and removes your personal workspace and your sign-in seven days later, unless you sign back in and cancel; "Delete your account" above says what it leaves. To have a shared organization removed as a whole, write to privacy@agentlefs.com; there is no self-service control for that yet.
The providers named above keep their own records for their own periods — session recordings and analytics under the retention settings configured in those services, sign-in records until your account is deleted. Ask us at privacy@agentlefs.com and we will tell you what applies at the time you ask.
How it is protected
Content is encrypted in transit, and stored with encryption at rest at our hosting platform's storage layer. Credentials for services you connect get a second layer on top of that: they are encrypted in the database under a separate key, and are never shown back to anyone, including us.
Access is decided per person and per document, and every read and write is recorded in a tamper-evident audit log your administrators can inspect.
Changes to this policy
If we change what we do with your data, we will change this page and update the date at the top. Where the change is significant, we will tell you in the product before it takes effect.
Contact
Questions about this policy, or about data we hold: privacy@agentlefs.com.

