TabAI Inc., a Delaware corporation ("TabAI," "we," "us," or "our"), operates Heka, the Heka macOS application, the website at useheka.com, and the paid data-contributor program connected to them (together, the "Service").

Heka records detailed computer activity during recording sessions that a participant starts and controls on the participant's own Mac, and anonymizes that activity on the device before anything uploads. A recording can still contain personal information belonging to the participant or another person. We use accepted recordings to build datasets, train and evaluate artificial-intelligence systems, and license datasets to companies that develop AI models and products.

Read this Privacy Policy before joining the waitlist, installing Heka, or starting a recording. Do not use Heka if you do not understand or agree with the practices described here.

This policy gives notice; accepting the Heka Terms does not by itself supply consent for every processing purpose. Where law requires consent, Heka requests it separately and records the version and time of the choice.

1. Scope

This Privacy Policy applies to:

people who visit our website, join our waitlist, apply to the contributor program, record with Heka, or contact us;
information that Heka records on a participant's Mac during a recording session;
people whose information appears in a submitted recording, even if they are not the participant; and
personal information that we process to operate, secure, and administer the Service.

This policy does not govern a Data Partner's independent products or later processing. Section 8 explains when a Data Partner receives an accepted Dataset and acts as a separate data controller or business.

2. Who Controls the Information

TabAI Inc. controls the personal information described in this policy, except where a Data Partner independently decides how and why to process information that it receives under a dataset license.

TabAI Inc.
670 2nd Street
San Francisco, CA 94107
United States
support@useheka.com

3. Information We Collect

3.1 Waitlist, application, and contact information

We may collect:

name, email address, country or region, preferred language, and contact details;
confirmation that the applicant is at least 18 years old;
waitlist answers, qualifications, device compatibility, availability, and recording preferences;
messages sent to support, feedback, survey responses, and records of consent;
reward terms, session submission history, acceptance or rejection decisions, and productivity and quality-review records; and
payment, tax, identity-verification, or sanctions-screening information when needed to pay a participant or comply with law.

We have not selected a payout, identity-verification, or tax vendor as of the effective date. We will update this policy and identify the relevant category of provider before using one.

3.2 Data recorded during a recording session

When the participant starts a Heka recording after granting the relevant macOS permissions, the App may collect the following information from the participant's personal Mac:

continuous screen video, still screenshots, display dimensions, visible cursor activity, and visual content shown on the recorded display;
mouse movement, clicks, dragging, scrolling, coordinates, timing, and related input events;
keyboard events, typed characters, hotkeys, and timing outside protected fields and blocked applications;
clipboard changes, clipboard text or supported rich-text content, content hashes, source-application information, and copy or paste timing;
application names, bundle identifiers, process identifiers, window titles, focus changes, and timestamps;
browser tab URLs, page titles, page structure, Document Object Model content, supported form values other than protected password values, and browser interaction data;
Accessibility information exposed by macOS, including interface roles, labels, descriptions, identifiers, values, positions, and the element under the pointer or keyboard focus;
file-system activity metadata for designated folders such as Desktop, Documents, and Downloads, including a file path, operation type, timestamp, and file size;
classifications produced from system-output audio, such as speech or music classifications and confidence scores;
participant-declared intent, completion or failure markers, and other participant-entered annotations; and
session identifiers, recording timestamps, chunk identifiers, censoring status, upload state, and integrity information.

File-system monitoring records metadata and does not intentionally read file contents. File contents can still appear if the participant displays them on screen, copies them to the clipboard, or opens them in a browser or application during recording.

Heka may use macOS Automation and Apple Events permission to obtain an active browser URL and page structure. Screen Recording, Accessibility, Input Monitoring, and Automation grant access beyond an ordinary app permission, so participants should grant them only after reading the in-App recording notice.

The current production design does not intentionally record microphone waveforms, camera video, address-book contacts, or precise GPS location. Heka may classify system-output audio without saving the audio waveform. We will update this policy and request any consent required by law before adding materially different capture features.

3.3 Device, network, and diagnostic information

We may collect:

a pseudonymous device identifier, participant identifier, and session identifiers;
macOS version, computer model, display configuration, keyboard layout, locale, time zone, language, and Heka version;
IP address and network information observed when the App connects to our systems;
CPU, memory, disk, and network performance information;
local storage usage, upload speed, upload progress, failure status, and censoring status; and
application logs, crash reports, stack traces, error messages, and security records.

The device identifier helps us connect recordings made by the same installation. Hashing or replacing a direct identifier does not necessarily make the information anonymous.

When program upload is enabled, Heka may periodically package and send diagnostic logs to Google Cloud under the pseudonymous device identifier. These log packages can include crash backtraces, resource measurements, local Heka folder sizes, censoring and upload status, network errors, App version, and technical identifiers. They should not intentionally contain uncensored screen video or raw input, but an error message or path can still contain personal information.

3.4 Information we create

TabAI, its service providers, and automated systems may create:

censored or replaced versions of recorded content;
optical-character-recognition results and detections of possible sensitive information;
session segments, labels, annotations, summaries, action descriptions, and productivity and quality measurements;
completion, validity, duplication, integrity, efficiency, and fraud indicators;
dataset indexes, embeddings, features, and other structured representations; and
model-training, evaluation, safety, and benchmarking results.

We call the censored recording and related data submitted to TabAI a "Contribution." Uncensored source material that remains only on the participant's Mac is not part of the submitted Contribution. Once TabAI accepts a Contribution and credits the rewards required by the applicable reward terms, we call the censored recording, screenshots, events, labels, metadata, and related processed material an "Accepted Dataset."

3.5 Information about other people

A screen recording can incidentally show another person's name, message, image, account, work product, health information, financial information, or other personal data. Participants must record only activity that they have the right to record, must not target another person's private information, and must pause recording or block the relevant application when such content appears. Heka never captures designated messaging applications. Participants must not intentionally submit information about another person without every consent that law requires.

If your information appears in a Heka recording that someone else submitted, you may contact us under Section 18.

4. How We Collect Information

We collect information:

directly from applicants and participants through forms, support messages, and consent screens;
automatically from the Heka App only while its recording features run;
from the participant's device and operating-system interfaces after the participant grants Screen Recording, Accessibility, Input Monitoring, Automation, or another permission described in the recording notice;
from service providers that host or process data for us; and
from Data Partners when they report dataset issues, safety concerns, deletion status, or license compliance.

The participant controls when a session starts and stops. Some diagnostic or upload activity may continue after recording stops so that Heka can censor, validate, upload, or troubleshoot an already-created Contribution.

5. Local Censoring and Upload

Heka applies privacy filters on the participant's Mac before TabAI receives a Contribution. Depending on the feature and context, those filters may:

suppress keystrokes and clipboard content in protected fields or blocked applications;
cover windows from blocked applications;
remove protected values from browser data;
detect and replace or cover likely secrets and personal information; and
quarantine a Contribution when censoring fails.

Privacy filters can miss information. Participants must supervise their recording environment, use the App's pause and blocking controls, and stop recording when sensitive material appears. A filter does not give permission to record information that the participant has no right to submit.

Uncensored source data stays on the participant's Mac and does not transfer to TabAI under the production Service. Heka may keep temporary local working files or backups while it censors, validates, recovers, or uploads a Contribution. Those local files remain under the participant's control and may contain sensitive information. The participant must protect the Mac and delete local files according to the App instructions.

Only a censored Contribution may upload to TabAI's Google Cloud environment. TabAI does not accept an uncensored upload through the production contributor program. Censoring reduces risk but does not necessarily make a Contribution anonymous.

6. Why We Process Information

We process personal information to:

operate the waitlist, assess eligibility, select participants, and communicate about the program;
operate recording sessions and assess the amount of productive work they contain;
record consent and enforce age and territory restrictions;
receive, censor, validate, review, accept, reject, and pay for Contributions;
create, organize, label, test, maintain, and improve datasets;
train, fine-tune, test, benchmark, evaluate, and improve TabAI's AI systems and commercial products;
license Accepted Datasets to Data Partners for their AI research, training, testing, evaluation, safety work, and commercial product development;
use Google Cloud and the enterprise Google AI service identified in the applicable provider or program notice to store, process, label, classify, or analyze censored Contributions;
detect fraud, fabricated sessions, duplicate submissions, malware, policy violations, and security incidents;
provide support, diagnose technical failures, and maintain the Service;
protect TabAI, participants, Data Partners, and other people;
keep financial, tax, audit, and legal records;
establish, exercise, or defend legal claims; and
create information that no longer identifies a person, then use that non-personal information for lawful purposes.

We do not currently use Heka Contributions for cross-context behavioral advertising.

7. Legal Bases for Processing

The legal basis depends on the information, purpose, and participant's location.

Contract

We process information needed to evaluate an application, operate recording sessions, review a Contribution, credit rewards, make payment, and enforce the Heka Terms. Taking steps at an applicant's request before entering an agreement may use the same basis.

We ask for affirmative consent before starting Heka's recording and upload features. The ordinary contributor program does not authorize recording designed to collect health information, race or ethnicity, religious or philosophical beliefs, political opinions, trade-union membership, sex life or sexual orientation, biometric identifiers, criminal-offence data, or other specially protected information.

If a separately approved project would intentionally process the participant's own specially protected information, we will provide a project-specific notice and request explicit consent where law permits that processing. General recording consent does not authorize a participant to submit specially protected information about someone else.

Where we rely on consent, a person may withdraw that consent for future processing. Withdrawal does not make earlier lawful processing unlawful. An accepted and paid Contribution also carries the contractual and intellectual-property terms described in the Heka Terms, subject to rights that applicable privacy law does not allow a person to waive.

Legitimate interests

Where allowed, we process information for our interests in securing the Service, preventing fraud, reviewing dataset quality, improving recording and censoring tools, handling support, protecting legal rights, and running the business. We balance those interests against the rights and expectations of affected people.

Legal obligations

We process information when tax, accounting, sanctions, employment, law-enforcement, court, regulatory, or other legal rules require it.

If another legal basis applies under local law, we will identify it in the relevant notice or request.

8. How We Disclose Information

8.1 Google service providers

We use Google Cloud to receive, store, secure, and process censored Contributions and related logs. For a project that needs automated labeling, we may use an enterprise Google AI service that we identify in a provider list or program notice. We do not treat consumer and enterprise Google AI products as interchangeable because their retention and provider-use terms can differ.

Before sending participant data to a Google AI service, we select the account type, region, settings, retention, and business terms for that project. We do not authorize Google to use Heka Contributions for unrelated advertising, general model training, or human review unless we first disclose that different use and establish a lawful basis.

8.2 Operational service providers

We may engage providers for hosting, email delivery, waitlist management, analytics, support, security, identity checks, sanctions screening, tax administration, and payouts. We currently use Supabase to store and operate the website waitlist, PostHog to provide website analytics and masked session replay, and Vercel Analytics to measure aggregate site traffic and performance. We do not send the email address entered in the waitlist form to PostHog.

These providers process information for the operational purpose for which TabAI engages them. They do not become Data Partners merely because they provide infrastructure or analytics services. We will update this policy or a linked provider list before materially changing these uses.

8.3 Data Partners and dataset buyers

TabAI licenses Accepted Datasets for monetary or other consideration to AI developers, model providers, research organizations, and other commercial partners ("Data Partners").

A Data Partner may receive the full censored, processed dataset covered by its license. Depending on the dataset, this may include screen video, screenshots, pointer and keyboard events, browser and interface data, session labels, annotations, technical metadata, and derived information. Data Partners may use the Dataset to train, fine-tune, test, benchmark, evaluate, secure, and develop commercial AI models and products.

Data Partners may receive personal information even after TabAI censors direct identifiers. They may act as independent controllers or businesses when they decide how to train or develop their own systems. They do not act solely as TabAI's service providers in that situation.

Before disclosure, we require a written agreement that limits the partner to authorized AI purposes and prohibits unauthorized reidentification, contact with recorded people, credential extraction, targeted advertising, biometric identification, public release of source recordings, unlawful surveillance, decisions about employment, credit, insurance, housing, education, health care, or similar eligibility, and unapproved onward transfer. We also require security, incident-reporting, retention, and deletion terms.

TabAI does not make source Contributions publicly available through the Service. No security or contractual restriction can eliminate every risk after disclosure.

8.4 Legal, safety, and corporate disclosures

We may disclose information:

to attorneys, accountants, auditors, insurers, and other professional advisers;
to courts, regulators, law enforcement, tax authorities, or other government bodies when law requires it or when needed to protect rights and safety;
to investigate fraud, security incidents, misuse, or violations of our terms;
in connection with financing, due diligence, a merger, acquisition, reorganization, bankruptcy, or sale of some or all of TabAI's assets; and
with a person's direction or separate consent.

9. When Dataset Licensing Is a Sale

TabAI receives value for licensing Accepted Datasets. When an applicable privacy law defines that exchange as a "sale," TabAI treats it as a sale even though TabAI does not use participant information for targeted advertising.

Where applicable law gives a person the right to opt out of a sale or sharing, the person may submit a request to support@useheka.com. An opt-out may make the person ineligible to submit future Contributions because dataset licensing forms a core part of the contributor program.

An opt-out applies prospectively unless law requires more. It does not automatically reverse a disclosure that occurred lawfully before TabAI received the request. We will notify Data Partners or take other steps when applicable law requires us to do so.

Where required, we treat a legally recognized browser opt-out signal, such as Global Privacy Control, as an opt-out for the browser and account that we can reasonably associate with the signal. Because Heka Contributions come from a desktop application, we may need additional information to connect a browser signal to a participant or Contribution.

9.1 Notice of financial incentive

The contributor program offers compensation in exchange for an Accepted Contribution and the license described in the Heka Terms. Depending on applicable law, this arrangement may qualify as a financial incentive or price difference related to personal information.

The reward terms in the App and on the website explain how TabAI determines rewards before the participant records. TabAI estimates the value of a Contribution using factors such as session duration, usable recording time, the amount of productive work shown, quality, complexity, rarity, collection and processing costs, expected dataset use, and the value that a buyer may place on the data. The compensation difference reasonably reflects those factors rather than a participant's protected status.

A person opts in by joining the program, separately consenting to recording, submitting the Contribution, and accepting rewards after TabAI approves it. A participant may pause or stop recording at any time and may stop before submission. The Heka Terms describe how a participant may request deletion of an uploaded recording; non-waivable privacy rights also apply.

To withdraw from future incentives, stop recording and email support@useheka.com. Withdrawal does not affect compensation already earned.

10. AI Processing and Quality Decisions

TabAI and Google AI services may produce labels, summaries, classifications, quality signals, and other model-generated information from a censored Contribution. Automated output can be wrong.

We may use automated signals to help identify corrupt, duplicate, fabricated, unsafe, incomplete, unproductive, or low-quality Contributions. If an automated score materially affects acceptance or rewards, the applicable reward terms or program notice will describe the relevant factors, such as the share of productive activity in a session. A participant may request human review of the result.

We do not make a final acceptance or payment decision that produces a legal or similarly significant effect solely through automated processing. A trained reviewer must have authority to assess the relevant evidence and change the result.

11. Retention

We keep information according to its purpose, sensitivity, legal requirements, and status:

Uncensored local data: TabAI does not receive it. Temporary files and backups may remain on the participant's Mac until Heka or the participant deletes them.
Rejected or abandoned Contributions: We delete or deidentify cloud intake copies within 90 days after rejection or abandonment, unless we need a limited copy to investigate fraud, security, payment, or a legal claim.
Accepted cloud intake copies: We delete temporary upload, staging, and redundant intake copies within 90 days after processing, subject to backup cycles and legal holds.
Accepted Datasets: After TabAI accepts and pays for a Contribution, the censored videos, screenshots, events, labels, metadata, and derived material that become part of an Accepted Dataset may remain longer than 90 days. We retain them while reasonably needed for the disclosed training, evaluation, licensing, safety, audit, and legal purposes. We review that need at least annually and delete or deidentify data that no longer serves those purposes.
Data Partner copies: A partner may retain an Accepted Dataset for the period stated in its written license and as law permits. Models, evaluations, statistics, and genuinely anonymous derivatives may remain after source-data deletion when they no longer identify a person.
Application, payment, tax, and legal records: We keep them for the period required by tax, accounting, sanctions, contract, and other legal rules.
Security, support, and diagnostic records: We keep them as long as needed to resolve the issue, prevent recurrence, protect the Service, or meet legal duties.

The 90-day intake limit does not mean that an accepted and paid recording disappears after 90 days. Its censored content can become part of an Accepted Dataset retained and licensed as described above.

We may suspend deletion when law requires preservation, a dispute remains open, or we need evidence to investigate misuse. We delete or deidentify the information when that reason ends.

12. Participant Controls

Heka provides the following recording and upload controls:

start and stop a recording;
block designated applications or windows;
enable or disable program upload;
prevent ordinary upload when the required censoring system is unavailable; and
view recording and upload status.

Local-file deletion is separate from these controls. The App instructions identify Heka's local storage location, and a participant may remove eligible local files when Heka is not actively processing or uploading them. A dedicated delete button may not be available in every App version.

Participants must use these controls and follow the in-App guidance. Removing an application from a blocklist or bypassing a privacy filter does not change the participant's duties.

A participant may stop future collection by stopping the App, revoking macOS permissions, or uninstalling Heka. These steps do not delete information that TabAI or a Data Partner already received.

A participant may request deletion of an uploaded recording as described in the Heka Terms. Deletion covers copies under TabAI's control, subject to the retention rules in Section 11; a dataset copy already delivered to a Data Partner is handled under that partner's written agreement and applicable law. This Section does not limit privacy rights that applicable law makes non-waivable.

13. Privacy Rights

Depending on location and applicable law, a person may have the right to:

ask whether we process their personal information and obtain access to it;
request correction, deletion, or restriction;
receive certain information in a portable format;
object to certain processing, including direct marketing or processing based on legitimate interests;
withdraw consent for future consent-based processing;
opt out of qualifying sales, sharing, targeted advertising, or some profiling;
limit the use or disclosure of certain sensitive personal information;
appeal our refusal of a request; and
complain to a privacy or data-protection regulator.

Submit a request to support@useheka.com. State the right you want to exercise and provide enough information for us to locate the relevant records. We may verify identity and authority before acting. If we cannot verify a request without collecting excessive additional data, we will explain the problem.

We respond within the period required by applicable law. This generally means one month for a qualifying EEA or UK request and 45 days for a qualifying request under many US state laws, subject to a lawful extension and notice.

An authorized agent may submit a request where law allows. We may ask for proof of authorization and direct identity confirmation from the affected person.

We will not unlawfully discriminate against a person for exercising a privacy right. A request may affect eligibility for the contributor program when the requested restriction conflicts with the program's core dataset purpose.

No contract can remove a right that applicable privacy law makes mandatory. At the same time, a deletion or withdrawal right may have exceptions, and privacy law may not apply to information that no longer identifies a person. We assess each request under the law that applies to it.

EEA, United Kingdom, and Switzerland

People in these locations may have rights of access, correction, erasure, restriction, portability, objection, and withdrawal of consent. They may complain to the data-protection authority where they live or work or where an alleged violation occurred.

If we deny a request based on an exception, we will identify the basis where law requires an explanation.

United States

Residents of California and other states may have rights to know, access, correct, delete, obtain a copy of, or opt out of certain processing of personal information. State law may also grant an appeal right and protections concerning sensitive data.

For California notice-at-collection purposes, the categories we may collect include identifiers; customer or contributor records; protected-classification information if it appears in a Contribution; commercial and payment records; internet or electronic-network activity; geolocation at a general IP-derived level; audio and visual information; professional or employment-related information; education information if submitted; inferences; and sensitive personal information. Sections 3, 6, 8, and 11 state the sources, purposes, recipients, and retention criteria.

We may disclose or license identifiers, internet or device activity, audio or visual information, professional information, inferences, and sensitive information that remains in an Accepted Dataset after censoring. We do not knowingly sell personal information belonging to anyone under 18.

Section 9.1 supplies notice about compensation that may qualify as a financial incentive. Eligible residents can opt out of future qualifying sales or sharing as described in Section 9.

To appeal our response to a state privacy request, reply to our decision or email support@useheka.com with the subject "Privacy Appeal."

14. International Transfers

TabAI operates from the United States. We and the recipients described in Section 8 may process information in the United States and other countries whose privacy laws differ from those in the participant's location.

Where law requires a transfer mechanism, we use contractual clauses approved by the relevant authority, a valid adequacy decision, an approved certification, or another lawful safeguard. We also assess technical and organizational protections appropriate to the transfer.

15. Security

We use controls intended to reduce the risk of unauthorized access, loss, misuse, or alteration. These controls include local censoring before upload, encrypted network transport, access restrictions, logging, review of service providers, and written restrictions for Data Partners.

Heka does not promise that censoring will find every secret or personal detail. We also cannot guarantee that storage or transmission will remain secure in every circumstance. Participants should use the App's pause and blocking controls, protect their Mac account, and stop recording whenever unexpected sensitive content appears.

If we identify a personal-data breach, we will investigate and notify affected people and regulators when law requires notice.

16. Adults Only

The Service and contributor program are for people aged 18 or older. We do not knowingly recruit participants under 18 or license their Contributions.

We may use a neutral age screen before collecting a Contribution. If we learn that a participant is under 18, we will stop their participation and delete or restrict their personal information as required by law. A false age statement violates the Heka Terms, but it does not remove rights or duties that law makes mandatory.

Recording may not target children's accounts, services, communications, or content. If we learn that a Contribution contains a minor's personal information, we may quarantine it, stop distribution, notify affected Data Partners, and delete or restrict it as law requires.

Contact support@useheka.com if you believe a minor submitted information through Heka.

17. Website Technologies and External Services

Our website uses technologies strictly needed to deliver pages, maintain security, remember a visitor's privacy choice, and operate forms. A local-storage entry remembers whether the visitor opted out of optional analytics.

We use PostHog to understand and improve the website. Unless a visitor opts out, PostHog may process:

page paths, referring domains, and campaign parameters;
browser, device, language, time-zone, network, and approximate IP-derived location information;
clicks, navigation choices, outbound-link use, FAQ interactions, section views, scroll depth, and active-time milestones;
waitlist-form start, validation, submission, success, or failure events, without the entered email address;
performance measurements, web vitals, dead or repeated clicks, and unhandled browser errors; and
a replay of the website session used to understand layout and interaction problems.

We configure replay to mask every form input and block elements marked as sensitive. We do not enable replay capture of request or response bodies, request or response headers, or console logs. We remove query strings and fragments from analytics event URLs and do not intentionally place passwords, access tokens, or waitlist email addresses in analytics properties.

PostHog uses first-party browser storage to connect events within and across visits. PostHog may receive the network IP address needed to deliver the service and derive an approximate location. Website analytics are processed in PostHog's United States cloud environment. Session replays are configured for a 30-day retention period; other analytics data follows the retention criteria in Section 11.

We also use Vercel Analytics, our hosting provider's aggregate measurement service, to count page views and measure site performance. Vercel Analytics does not use cookies or persistent browser identifiers.

The pre-interview form that our invitation emails link to does not use PostHog. Instead we keep our own record of how the form is used, stored with Supabase: each time the page loads and what it showed, the approximate location our hosting provider derives from the IP address, the device type, operating system, and browser, which fields were selected and for how long, whether each field ended up filled, the multiple-choice answers picked (such as occupation, computer, and hours a day at a computer), any error messages shown, and whether the form was sent. We do not record what is typed into the fields. Each invitation link is personal, so these records are tied to the email address we sent it to. We use them to improve the form and the invitation, and they follow the retention criteria in Section 11.

A visitor can opt out of optional analytics through Privacy choices in the website footer by selecting Only essential, which prevents PostHog analytics, custom events, heatmaps, and session replay from starting and stops that browser from sending the pre-interview form records. A recognized Global Privacy Control or Do Not Track signal also keeps optional analytics off, including every pre-interview form record. The choice applies to future activity; opting out does not delete information already processed, and a deletion request may be sent to support@useheka.com.

We do not use PostHog for third-party advertising, cross-site tracking, or targeted advertising.

The Service may link to websites or products that TabAI does not control. Their privacy practices govern information that a person gives directly to them.

18. Changes and Contact

We may update this Privacy Policy when Heka, our data practices, or applicable law changes. We will change the date at the top and give additional notice of a material change when law requires it. We will obtain new consent before applying a materially different consent-based purpose where applicable law requires new consent.

Questions, requests, complaints, and security reports may be sent to: