Privacy Policy

This policy explains what personal data Claritty collects, how we use it, who we share it with, and your rights (including under the GDPR and the CCPA). It applies to the Claritty platform at claritty.ai and app.claritty.ai.

Where you use Claritty for your own purposes, we are the controller of your account data. Where Claritty processes content you connect to do a job for you — your documents, your customers’ messages — you are the controller and we act as your processor, handling that content only to carry out the work you ask for.

Data we process

• Account data — your name, email, and organization, provided when you sign up. Sign-in is handled by Auth0.
• Content you give us — the prompts and instructions you write, the apps, automations and agents you build, and the jobs you hand over.
• Connected content — data we read from tools you connect (for example a Notion page, a Slack thread, a Drive document) in order to do the work you asked for, including content we index so it can be searched later.
• Credentials for connected tools — OAuth tokens and API keys, stored encrypted (see below).
• Usage data — logs, token consumption, run history, and errors, used to operate, meter, secure, and debug the service.
• Billing data — processed by Stripe. We do not store full card numbers.

How AI processing works

To do the work you ask for, we send the relevant content — your prompt, and the context the job needs — to the model provider that runs your plan’s model. Today those providers are Anthropic and OpenAI, and we use AWS Bedrock to turn indexed documents into embeddings so they can be searched.

We do not train foundation models on your content, and we do not permit our model providers to train on it. Content is sent to a provider to produce a result for you, and for no other purpose. If you bring your own model key (Team and Enterprise), those calls run under your own agreement with that provider and are governed by their terms.

AI output is generated, and may be wrong. We keep a record of what your agent did — the run log, the tools it called, the actions it took — so that you can audit it.

Connected tools and your credentials

When you connect a tool, we store the resulting credential encrypted at rest (envelope-encrypted with a managed key) and scoped to you and the app you connected it for. Credentials are never exposed to the AI model and are never shared with other users: every call to a connected provider is brokered server-side on your behalf. Disconnecting a tool deletes the stored credential and stops any future access.

Content read from a connected tool is processed to carry out your instructions, and — where you enable knowledge indexing — stored so it can be searched later. You choose which sources are indexed, and you can remove them.

The Claritty browser extension: recording

Claritty is a browser extension you install yourself, and it does two separate things: it records a task you show it, and it reads a thing you point at. They are described separately here because they behave differently, and a single sentence covering both would be a half-truth about each. This half is recording.

It is dormant until you press Record, and it records only the tab you started in and tabs opened from it — never every tab you have open.

While you are recording, it captures the STRUCTURE of what you do rather than the content you do it with:

• What you interacted with — the visible label, role and position of the control you clicked or typed into (“the Submit button”, “the Vendor column of a 5-column table”), the page title and path, and the shape of the region around it. The values you type are masked in the extension, before anything is sent.
• Screenshots of the tab, taken at a few key moments so the task can be understood from what the page actually looked like. These are stored encrypted and scoped to you. This is the most revealing thing the extension captures, which is why the recorder says so on screen the whole time it is recording.
• The address of each page you visit while recording, whole — including anything after the ?. That address is how the automation finds the page again: a list you filtered down to one batch is a different address from the unfiltered one, and without it the automation would open the unfiltered one and act on everything. A parameter carrying a token, a password or a session is dropped from it before the address leaves your browser.
• The shape of network requests the page made — method, host, and path with identifiers removed (/api/invoices/{id}), status and content type. This is how Claritty discovers that a system has an API and can skip driving a browser. No request or response bodies, no headers, no query values, and no authentication of any kind is read or stored.
• Anything you type into the recorder yourself — the goal you describe and the notes you add to a step. This is the one field you fill in, so treat it as you would any message to us: it is sent to a model, and personal details in it are scrubbed on arrival.

What a recording never captures: the values in form fields, the text of the records on a page, your passwords, or the contents of any page while it is not recording. While it is recording, it reads one cookie — your Claritty session — so that it can talk to your own account, and no others. Connecting a site is the one other time it reads cookies, and it is described next.

That paragraph is about recording and only about recording. Pointing at something is the other gesture, it does read what is on a page — that is what you asked it to do — and it has its own section below. The two used to be two extensions, which is why you may still have a separate one called Claritty Ask installed; the section below describes it too.

A recording is processed by a model to turn it into an automation, on the same terms as everything else in this policy, and is kept with the automation it produced. Deleting the automation deletes the recording behind it. Uninstalling the extension stops all capture immediately.

The Claritty browser extension: pointing at something

The other gesture, in the same extension: you hold a key and click something on a page, and it tells you what it read and answers questions about it. Because that is the job, it does read page content — and every one of the limits below is in the extension rather than in this promise.

You are asked before the first time, in the browser, because a recorder that never read a page and an extension that reads what you point at are not the same bargain. Nothing is read by this gesture until you say yes to it once.

This was a second extension of its own, called Claritty Ask, and some people still have that one installed separately. Everything in this section describes it as well.

Nothing is read until you point at something. There is no background capture and no reading of pages you are not asking about. The gesture is deliberate: a held key and a click, or the extension’s own keyboard shortcut.

Naming what you pointed at happens on your machine. The role, the accessible name and the column headers around it are read locally, by the same reader a recording uses, and that reading never leaves the browser. It is why the extension can tell you it is looking at the “MRR” column while you are signed out and while nothing has been sent.

A site has to be agreed to, once. The first time anything would leave a website, the panel asks whether to use Claritty there and remembers your answer. Sites you have not agreed to send nothing.

What leaves when you ask: the question you typed or spoke, the page’s address and title, what you pointed at — and the contents of that region: up to 20 rows, 12 columns, 120 characters a cell. A question about data cannot be answered without the data, and a plan has to fill in a real recipient and a real summary; a list of column names is neither.

When the thing you pointed at is words, the words go too. A post, a message in a list, a paragraph of terms, or the text you highlighted before you asked — up to about four thousand characters of it, redacted the same way the rows are. It is the region you pointed at and not the page around it: if you highlight one sentence, that sentence is what is sent.

Handing something to an agent sends the same content once more, into that task, where it stays as long as the task does. An agent has no browser and cannot open your page, so a task that only carried a link would be a task nobody could do.

Those rows arrive redacted. A column whose header names something sensitive — an email address, a card, a national ID — is withheld whole. An address or a card number sitting in a column that does not announce itself is caught by pattern. A withheld value is replaced by a marker saying a real value is there and was not read, so nothing plans around a cell it thinks is empty.

A screenshot is taken only when there is nothing else to read — a canvas grid, a chart, a PDF, an embedded viewer — and only of the region you pointed at, cropped in the extension before it is sent. On a PDF, where Chrome’s viewer takes the mouse and there is nothing to point at, that region is the page you are looking at. The fields a site itself marks sensitive are blanked before the image exists, and no image is taken at all while a password field is in use.

One cookie: your Claritty session on app.claritty.ai, so it can talk to your own account and no others. It reads no cookies for the sites you browse. And it refuses to touch a page while a recording is running on it.

Spoken questions are transcribed on your machine when the browser can do it, and by Google when it cannot — the panel says which, every time it is listening.

Asking whether a site can be connected reaches that site. On a site we have no integration for, the panel offers to look for one. Pressing it asks Claritty to fetch the few addresses a published standard says an API description lives at, on that site — nothing else is requested, no credential is sent, and it happens only on that press. A page you merely looked at is never fetched.

Talking to Claritty

Home has a Talk to Claritty button. When you press it, Claritty reads out what needs you and what happened, using your browser's own voice on your device. When it listens, your browser turns your speech into text: Chrome does this by sending the audio to Google, under Google's terms. Claritty receives only the words, never the audio, and keeps no recording.

What you say is used for what you asked: answering one of Claritty's questions, opening a job, or going into the box on Home as if you had typed it. Speaking never sends, approves or changes anything by itself; that still takes a tap.

Connecting a site to an automation

An automation that works a site behind a login has to be signed in to that site when it runs. Rather than ask you for the password, Claritty borrows the session you already have: you press connect this site in the extension while you are on it and signed in, and nothing happens until you do.

What is taken, when you press it: all of that site’s cookies for that host, and the values that page keeps in browser local storage — because that is where many sites hold the sign-in. This is more than the single cookie a recording reads, and it is the reason the extension asks for cookie access at all.

It is deliberately a session and not a password. A session expires on its own, it can be revoked by signing out of that site, and it is the narrowest thing that makes the automation work. We never ask for, receive, or store the password itself.

How it is held: encrypted with a key held in AWS KMS, and scoped to one user, one automation and one site. Another automation of yours cannot use it, and no other customer can resolve it. It is loaded into a Claritty-controlled browser at the moment of a run and is never written to disk there. Deleting the automation, or removing the connection from it, deletes the stored session.

Sites for which we have a real integration — Gmail, Slack, HubSpot and the rest — never use this path. They use that provider’s own authorization, described elsewhere in this policy.

Google user data (Gmail and other Google services)

When you connect a Google account to Claritty (for example, Gmail), you grant Claritty access to specific Google data through Google’s OAuth consent screen. We access this data only after you explicitly connect the integration, and only to provide the features of the Claritty app you are using.

What we access. We request the narrowest scope each app needs to function, and never more than the app you are using actually requires. Today Claritty requests permission to send email on your behalf (Gmail), to read and create calendar events (Google Calendar), and to open files you have specifically selected or that Claritty itself created (Google Drive).

Claritty does not currently request permission to read, search, label, archive or delete the messages in your mailbox. Those permissions require separate approval from Google, and if we ever offer them you would be asked to reconnect the integration and approve them yourself on Google’s consent screen. The consent screen you see when connecting is always the authoritative list of what you are granting.

How we use it. Google user data is used solely to perform the actions you or your connected app initiate (such as sending a message you approve, or adding an event to your calendar). We do not use Google user data for advertising, and we do not sell it.

How we store it. Your OAuth tokens are encrypted at rest (envelope-encrypted via a managed key) and held on the Claritty platform. Tokens are never exposed to the individual apps that use the integration — each Google API call is brokered by the platform on your behalf. You can disconnect an integration at any time, which revokes Claritty’s access and deletes the stored tokens.

How we share it. We do not transfer Google user data to third parties except as necessary to provide and secure the service (for example, our cloud infrastructure providers acting as processors under contract), to comply with applicable law, or as part of the operation you explicitly requested.

Limited Use. Claritty’s use and transfer to any other app of information received from Google APIs will adhere to the Google API Services User Data Policy, including the Limited Use requirements. We do not allow humans to read your Google user data unless (a) you give explicit consent to access specific items, (b) it is necessary for security purposes (such as investigating abuse), (c) it is required to comply with applicable law, or (d) the data is aggregated and anonymized and used to improve the service in line with that policy. Google user data is never used to train AI models.

Who we share data with

We do not sell your personal data. We share it only with providers who process it on our behalf, under contract, to run the service:

• Anthropic, OpenAI, and AWS (Bedrock) — to generate AI output and embeddings for you.
• Amazon Web Services — hosting, databases, storage, and key management.
• Vercel — hosting of the web application.
• Auth0 — authentication and sign-in.
• Stripe — payments, subscriptions, and creator payouts.
• Tools you connect — when Claritty acts in them at your instruction.

We may also disclose data where required by law, to protect our rights or the safety of users, or in connection with a merger or acquisition (in which case this policy continues to apply until you are told otherwise).

Where data is processed

We process data in the United States. If you are in the EEA or the UK, this means your data is transferred internationally; we rely on the European Commission’s Standard Contractual Clauses and equivalent safeguards for those transfers.

How long we keep it

We keep your account data for as long as your account is open. Content, run history, and indexed knowledge are kept until you delete them or close your account. When you close an account we delete or de-identify your data within a reasonable period, except where we must keep records (for example invoices, or logs held for security and fraud prevention). Credentials for a connected tool are deleted when you disconnect it.

Security

We encrypt data in transit and at rest, hold connected-tool credentials under envelope encryption with a managed key, isolate every workspace’s data, and restrict internal access to those who need it to run the service. No system is perfectly secure, and we cannot guarantee absolute security.

Your rights

Depending on where you live, you may have the right to access, correct, export, or delete your personal data, to object to or restrict processing, and to withdraw consent. You can disconnect any tool at any time from your integration settings — for Google, also from your Google Account permissions. To exercise any other right, contact us and we will respond within the time the law allows. You also have the right to complain to your local data protection authority.

We do not sell or share personal information for cross-context behavioural advertising, as those terms are used under the CCPA.

Children

Claritty is not intended for anyone under 18, and we do not knowingly collect data from children.

Changes and contact

We will update this policy as the product changes, and will tell you when a change is material. Questions, or to exercise a right: privacy@claritty.ai.