ErezAgent -- Privacy Policy

**Last updated: v4.94.5 -- 2026-10-18**

---

## 1. Who This Applies To

This policy covers two different relationships, and they are deliberately
kept separate:

- **You, the customer, interacting with Licensor directly** -- e.g.,
  emailing about a license purchase, contacting support, or visiting a
  Licensor-operated website. For this relationship, Licensor is a normal
  data controller under GDPR/CCPA, and this policy describes that.
- **The operational data inside your ErezAgent deployment** -- your
  missions, agents, financial records, customer data, conversation history,
  or anything else stored in your own database. Licensor does not act as a
  data controller or processor for this data, because it never reaches
  Licensor's infrastructure. See Section 3.

## 2. Data Licensor Collects Directly From You

If you contact Licensor (email, a purchase inquiry, a support request), we
collect what you send us: your name, email address, and the content of your
message -- including, if you choose to include it, your machine fingerprint
when requesting a paid license. This is used only to respond to you, issue
and track your license, and provide support. It is not sold or shared with
third parties, and is retained only as long as reasonably needed for those
purposes (e.g., ongoing license/support records).

If a future hosted or managed offering is introduced (see Section 4), this
policy will be updated before that offering collects any additional data.

## 3. What ErezAgent (the Software) Does *Not* Send Us

This section describes the actual behavior of the software as shipped,
verified directly against the source code as of this version:

- **Nothing is sent to Licensor.** ErezAgent does not collect analytics
  and does not transmit usage data, error reports, or operational data from
  your deployment to Licensor. A deployment can optionally be connected to a
  provider you choose (Section 4); that connection does not exist until your
  owner enters the provider's enrolment string and gives consent in
  Settings -> Provider / Managed Service.
- **License validation is offline.** Your license key is verified locally
  inside your deployment using an embedded public key (Ed25519 signature
  verification), directly or through the certificate chain carried inside the
  license file. No network call to a license server is made to validate it.
- **Outbound calls the backend makes on its own** are limited to: an
  anonymous `GET` request to GitHub's public releases API to check whether a
  newer version is available; a daily request to a public exchange-rate
  service for currency rates; and, for the AI providers you configure, status
  checks sent to those providers with your own keys. None of them carries your
  business data, and none goes to Licensor. They are in addition to the
  connectors and processors you configure yourself.
- **Machine fingerprint and trial data stay local.** When you activate a
  trial or bind a license to a machine, that fingerprint is stored in your
  own database, not sent anywhere automatically. It only reaches Licensor if
  you choose to email it when requesting a license (see Section 2).
- **Billing is opt-in and processor-of-your-choice.** If you enable Plan &
  Billing and connect Stripe, Rapyd, PayPal, or Pelecard, payment data goes
  directly to that processor -- the same pattern as any other connector you
  choose to activate. Licensor never receives or stores your payment details.
  The same is true if you enable a venture's own Venture Commerce subscription
  product (v4.77.0) -- that venture's own connected Stripe or PayPal account
  processes its own customers' payments directly; Licensor never sees them.
  Gateway credentials you enter in Billing Setup are stored encrypted in your
  own database and are never shown again or sent to Licensor.
- **Billing documents and usage reports stay with you (v4.89.0).** Usage
  statements, invoice drafts and the PDF or DOCX billing documents made from
  them are generated and stored inside your own deployment and contain the
  customer names or references you enter; an email you choose to send goes
  through the connector you configure. A data export or erase request for a
  person covers these documents; the underlying financial records are kept as
  financial records. If your vendor bills you by usage, you can export a usage
  report for one month: it holds aggregate figures only (no names or other
  personal data), a checksum and your attestation, and it is never sent
  automatically -- you decide whether to send it. Webhooks and exports for your
  own systems go only to endpoints you configure.
- **LLM provider spend/balance lookups are opt-in (v4.77.0).** If you enter a
  separate Admin/Management API key for Anthropic, OpenAI, or xAI (never your
  ordinary inference key) to see real spend or balance on the AI Resources
  page, the backend calls that provider's own API directly, using your key --
  the same opt-in, processor-of-your-choice pattern as billing above. Nothing
  is routed through, or visible to, Licensor.
- **Document Collection and WhatsApp are opt-in (v4.78.0).** If you connect
  the WhatsApp channel, inbound messages and media are fetched directly from
  Meta's API using your own connected account -- Licensor never sees them. A
  document your customer submits (by email, WhatsApp, or your own customer
  portal) is stored as-is inside your own deployment (on the same local vault
  storage your Obsidian integration already uses), never sent to Licensor,
  and never indexed into semantic search or AI memory, since it may contain
  personal or identifying information about your customer.
- **E-signature is opt-in and DocuSign-of-your-choice (v4.82.0).** If you use
  a Business Record's signature field to request a signature, the document
  and recipient details you send go directly to DocuSign using your own
  connected DocuSign account -- Licensor never receives or stores them.
  DocuSign's own privacy policy governs how it handles that data and the
  signer's consent; the status DocuSign reports back (sent, delivered,
  signed, declined) is stored inside your own deployment, same as any other
  Business Record data.
- **AI providers and voice are ones you configure.** When agents or chat run,
  the prompt content (which can include data from your deployment) is sent
  directly from your deployment to the AI provider(s) whose API keys you
  configure -- the catalog covers Anthropic, OpenAI, Google, DeepSeek, Mistral,
  and xAI -- under your own account and that provider's own terms and privacy
  policy. Voice features likewise call the speech provider you enable (OpenAI
  for speech-to-text; ElevenLabs, OpenAI, or a self-hosted Voicebox for
  text-to-speech). Browser push notifications are delivered through the push
  service of the user's own browser. Licensor receives none of this.
- **Venture staff accounts (v4.84.0)** and **Business Record customer
  portals (v4.85.0)** are additional ways *inside your own deployment* for
  people to reach data you already store there -- neither sends anything to
  Licensor. A venture staff account you invite can read and write only the
  one venture it's scoped to. A customer reaching their own record's portal
  link can see and act on only that one record, limited to the fields and
  actions you explicitly configure for that record type. Both are governed
  the same way as every other role in Section 1 -- data stays inside your
  deployment; who inside or outside your organization can reach which slice
  of it is something you control.
- **AI analysis of your business documents is opt-in (v4.88.0).** The Build
  My Business feature and the document-reading features of Document
  Collection and Business Records can send the text (or, for scanned pages
  and images, the page image) of documents you upload to an AI provider so it
  can be read. This stays off until the account owner turns on AI-analysis
  consent in Settings -> Data & Privacy, and the owner can withdraw it at any
  time. In the default mode, identifiers such as Israeli ID numbers, IBANs,
  payment card numbers, phone numbers, email addresses and secrets (API keys,
  tokens, private keys) are masked in the text before it leaves your
  deployment (a page image itself cannot be masked). In the optional local-only mode, only locally hosted or
  privacy-safe models are used, images are read only by a local vision model,
  and nothing is sent to an external provider. Features you start yourself on
  your own documents (for example customer requirements intake, contract
  analysis, knowledge generation and document ingest) do not depend on that
  consent: they always mask identifiers in the text first, and in the
  local-only mode they use a local model or stop. Documents you attach to a
  chat message are masked the same way before they reach an AI model (the text
  you type yourself is not changed), and in local-only mode attachments are
  refused. Other countries' ID formats, bank account numbers that are not
  IBANs, personal names and free-text addresses are not detected. The provider you configured receives that
  content under your own account and its own terms; Licensor receives none of
  it. Uploaded files and their extracted text are kept in your own
  deployment's storage and are deleted automatically once the retention
  period you set has passed (90 days by default; 0 keeps them), or sooner if
  you delete the bundle.

Because of this, Licensor is not a GDPR Data Processor or Data Controller,
and does not receive CCPA "personal information," with respect to whatever
data you choose to store inside your own ErezAgent deployment. That data's
privacy and security is governed entirely by you -- see `EULA.md` Section 5.

## 4. Hosted Operation and Provider Connection

ErezAgent is software that you run on your own infrastructure; Licensor does
not host your data. If a provider or MSP operates an ErezAgent install for you
under the Provider plan, that provider hosts and is responsible for that
deployment and for the data in it under a separate arrangement between you and that provider.
In that case, review the provider's own privacy terms. As of this version,
Licensor itself offers no hosted service.

**Optional provider connection (v4.94.0).** If your provider enrols your
installation, your owner can connect it to the provider in Settings -> Provider /
Managed Service. The connection is off by default, needs the owner's explicit
consent, and can be paused or disconnected at any time. While connected, the
installation reports to that provider: the ErezAgent version and migration
head, the number of workspaces, core-service status, a count-level security
summary, the status of AI and voice providers, connector counts, readiness
state, resource flags, and licence and subscription state. It does not send
prompts, model responses, business data, documents, e-mail, CRM data, names,
credentials, API keys or secrets. The page shows the exact last report. The
provider can reply with version information and notices; nothing in the reply
changes your installation. Your provider is responsible for the data it
receives; Licensor does not receive it unless Licensor is that provider.

## 5. Cookies and Local Storage

The ErezAgent web interface uses a single `httpOnly` session cookie for
authentication, scoped to your own deployment's domain. It is not a
third-party or tracking cookie. No cookies are set by, or shared with,
Licensor's infrastructure, because the frontend and backend both run inside
your own deployment.

## 6. Your Rights

Because your operational data lives entirely in your own deployment,
requests to access, correct, or delete that data (e.g., a GDPR Subject
Access Request from one of your own end users) are yours to fulfil as the
data controller. As of v3.104.0, Settings -> Data & Privacy (owner-only)
gives you a real self-service tool for this: enter a contact's email (and
optionally a phone number) to anonymize it across the tables that hold
personal data -- contacts, leads, support cases, appointments, documents,
business records, invoices and similar -- or download a structured export of
those tables as a ZIP -- alongside the direct database and admin access you
already have. Since v4.88.0 the list of covered tables is maintained in one
place and checked automatically, so a new table that holds personal data is
included.
Inside your deployment, a daily retention job removes system events after
90 days, audit logs after 730 days, operational log tables (error and notification
logs after 90 days; sign-in, security and governance audit rows after 730 days;
model-call, token and skill-invocation records after 400 days; workflow, task and
conversation records after 1 year; chronicle events after 3 years; the first
week after an upgrade only reports what it would remove), email send/delivery records after 1 year,
dismissed or rejected attention signals and failed sign-in records for unknown usernames after 30 and 90 days, and anonymizes the
contact email on closed deals after 6 months (closed lost) or 1 year (closed
won); it runs only after a recent completed backup. Other records are kept until
you remove them. The audit log records governed actions, including each time
a staff member or owner downloads a customer-submitted document file. The
windows for events and audit logs are configurable by you.
For data Licensor holds directly about you (Section 2), contact
erez.rozenbaum@gmail.com to request access, correction, or deletion.

## 7. Changes to This Policy

Material changes will be noted in `CHANGELOG.md`. Continued use of ErezAgent
or continued contact with Licensor after an update constitutes acceptance of
the revised policy.

## 8. Contact

Privacy questions: erez.rozenbaum@gmail.com