Privacy notice

This notice explains what information eva collects, why it is used, where service providers may process it, and the choices available to you.

Draft updated 29 October 2026 — subject to legal review before release

Who operates eva

EVA CEO operates the eva website, waitlist, and early-access service from Australia. Exact legal-entity, trading-name, ABN, and address details must be supplied and legally reviewed before public launch.

Information we collect

When you join the waitlist, we collect your email address, the date and source of the request, the consent-policy version shown to you, confirmation and unsubscribe status, and delivery records. We also process a one-way hash derived from your IP address to limit abuse; the raw address is not stored in the waitlist table.

If you later use eva, we may process account details and the business information you choose to add, including project context, documents, customers, assets, conversations, work items, and operating memory.

How information is collected and used

We collect information directly from forms and product interactions, and automatically from essential service logs and privacy-conscious performance or traffic measurement. We use it to administer the waitlist, send early-access invitations and updates, provide and secure the service, prevent abuse, respond to support and data requests, understand site reliability, and comply with legal obligations.

For signed-in users, eva records essential operational activity needed for support, reliability, security, and product improvement. This includes the account and verified project involved, a Supabase session identifier, a normalized route and feature, action or request outcome, duration, stable error code and one-way fingerprint, request and release identifiers, normalized browser, operating-system and device categories, and a monthly rotating one-way network hash. The raw IP address and full user-agent are not stored in eva's telemetry tables. Country remains uncollected unless a trusted edge-derived country signal is introduced.

Eva does not put request or response bodies, cookies, authorization headers, tokens, document or customer content, raw error messages, full stack traces, raw IP addresses, or full user-agent strings into product telemetry. Page views are kept separate from meaningful actions when understanding product use. Signed-in users can see their own 30-day activity summary; authorised platform administrators can use named activity and sanitized diagnostics to respond to support issues and investigate reliability. This essential operational collection does not have an in-product opt-out.

Eva also retains the existing anonymous daily product totals. Those totals do not store account or project identifiers, workspace content, IP addresses, or individual event histories. We assess the risk of identifying people from small totals before treating results as anonymous. Operational telemetry is separate from any future use of workspace content to train general models.

Optional Google connections

If you choose to connect a Google account, eva requests access to the Google services you select. Depending on the connection, eva may process Gmail messages and settings, Google Drive file information and selected content, Search Console properties and search-performance data, or Google Ads account details and keyword ideas. We use this information to provide the connected features within your selected eva project. Google account credentials and access tokens are handled by eva's integrations service and are not provided to the agent model.

For Keyword Planner research, eva sends the seed phrases, location, and language you select to Google Ads and receives estimated keyword ideas and demand. Search Console supplies information about searches where a connected website appeared. These connections are optional; eva's baseline content research can use your own phrases and questions without them. You can disconnect a Google account in eva's Integrations settings and contact us about access to or deletion of information already held by eva, subject to the retention exceptions below.

Other optional connections

If you connect another service, such as a Microsoft mailbox or files, a CRM such as HubSpot or Zoho, accounting software such as Xero or QuickBooks, or a bank feed through Basiq, eva processes the information that connection provides for the features you use in your selected project, and sends actions you approve to that service. That service handles the information under its own terms. Credentials and access tokens are handled by eva's integrations service and are not provided to the agent model. You can disconnect a service in eva's Integrations settings.

Planned private default and agent choice

Our product direction is a private default AI that keeps business data and intellectual property within a business-controlled environment, with the option to choose another agent. These capabilities are planned and are not available in the current build. Before introducing them, we need to describe the hosting and access arrangements, which features use private processing, retention, and any data shared with connected services or an alternative agent. This notice describes current processing below; the planned default does not change it today.

AI processing, training, and retention

eva uses OpenAI APIs for AI features by default, including dictation. Your requests and relevant business context, such as document excerpts, customer information, retrieved records, or conversation content, may be sent to OpenAI to generate answers and perform supported work. Voice conversations use ElevenLabs for speech recognition and spoken replies, and phone calls are carried by Twilio; those providers process the audio and call details needed for the conversation. Document generation runs in isolated execution sandboxes provided by E2B, which process the content needed for that job. Project access controls govern what eva retrieves; they do not mean that content is processed only within eva’s own infrastructure.

In a chat, you can choose which AI model answers. OpenAI remains the default, and a project owner can limit which models the project allows. The model you choose receives your messages and the business context eva retrieves for that conversation. Other AI features, such as dictation, document indexing and background work, keep using OpenAI. eva's servers choose the account and project for every request, and no model receives eva's or your provider credentials.

• Anthropic. When you choose a Claude model, your requests and relevant business context are sent to Anthropic. Anthropic states that it does not use inputs or outputs from its commercial API to train its models by default, and that it deletes API inputs and outputs from its systems within 30 days, except content flagged under its Usage Policy (kept for up to two years) or where the law requires longer. eva does not claim zero data retention with Anthropic.
• OpenRouter. When you choose a model shown as provided "via OpenRouter", such as a Google, Mistral or open-weights model, your requests and relevant business context are sent to OpenRouter, a model-routing service. OpenRouter forwards each request to a provider that runs the model, and the provider can differ from one request to the next. OpenRouter states that it does not store prompts or responses by default; it keeps request metadata such as token counts and timing. For every request, eva instructs OpenRouter to use only provider endpoints with a zero data retention policy and to exclude providers that may store or train on your data. If no provider meets those rules, the request fails instead of going elsewhere. These are OpenRouter's published policies and routing rules; the provider that runs the model handles the request under its own terms.

Before eva sends your content to any other AI provider, model-routing service, or AI provider account you connect yourself, this notice will name it and describe how it handles data, including whether it may use or retain your content.

OpenAI’s published API data policy states that API data is excluded from model training by default unless the API customer explicitly opts in. This is a provider policy, separate from eva’s use of project memory to support your work. Retention is separate from training: provider abuse-monitoring logs and some API features may retain content under their applicable terms and settings. eva does not claim zero data retention with OpenAI.

Public web research sends search terms or requested URLs outside eva. Connected actions may send the approved content to the selected service. These are distinct from keeping project records separate inside eva; do not assume that cloud hosting means data never leaves the platform.

Service providers and international processing

eva uses service providers for hosting, database and authentication infrastructure, website analytics, AI processing, and transactional email. These currently include Cloudflare, Railway, Supabase, OpenAI, and Resend. When you use the related features, ElevenLabs (voice), Twilio (phone calls), E2B (document execution), and Anthropic or OpenRouter (when you choose one of their models in a chat) also process information, and Stripe processes payment and billing details when you subscribe; eva does not receive or store your full card details. These providers may process information in countries outside Australia under their own security, privacy, and contractual controls.

We use Cloudflare and Railway service logs and HTTP and service metrics to operate the site and service. Application audit records help us investigate access and changes.

The eva.ceo website and app use Cloudflare Web Analytics to count page views and measure page performance. When a page loads, a Cloudflare script reports the page address without its query string, the referring site, and browser, device, country, and timing details. It sets no cookies, uses no browser storage, and does not fingerprint visitors. Visitors in the European Union are not measured. Cloudflare processes these measurements under its own privacy policy.

Retention and deletion

Waitlist requests are kept until you are invited, you join, or you unsubscribe. Invitation links expire after seven days. Rate-limit records are deleted after seven days. Joined and unsubscribed records are retained for as long as reasonably needed to administer the waitlist, honour suppression requests until an address rejoins, evidence consent, resolve disputes, and meet legal obligations.

Detailed operational telemetry is deleted after 30 days. Named daily usage summaries are deleted after 13 months. Final account disposal deletes that account's detailed telemetry and named usage summaries. Aggregate reliability and assessed de-identified product totals may remain because they do not contain account identifiers or business content.

When an account is closed, access ends immediately and cannot be restored. Projects still owned by that account close for all members. Their data stays inaccessible for 30 days and is then scheduled for disposal, including files and agent execution state. Work in projects previously transferred to another owner remains with those projects. If an editor closes an account, their access ends while the team project remains active. A documented legal hold may delay disposal when required for a lawful purpose; restricted backups may expire on a separate published schedule. Detailed project audit records and project-linked admin preview records are removed when a project is disposed. Account-linked admin audit records are removed when the account is disposed; the retention period for other admin audit records requires legal review.

Authorised platform administrators may read account and project content, audit events, and agent runs to operate and secure the service. A read-only preview can show the actual content available to a user. Every preview and sensitive record view is audited. Administrators cannot edit project content or act as the user, but can lock, unlock, or remove accounts under controlled procedures.

Your choices and rights

You can unsubscribe from waitlist updates using the link in an email, or by contacting us. You may also ask to access, correct, or delete personal information, or raise a privacy concern. Some information may need to be retained to meet legal requirements or ensure an unsubscribed address remains suppressed.

Contact and complaints

The public privacy contact address is awaiting operator confirmation. Email contact@eva.ceo.

If you are not satisfied with our response, you may contact the Office of the Australian Information Commissioner.