# Privacy Policy

AIsh.li respects your privacy and is committed to protecting your personal data and ensuring transparency.

_Last updated: [November 15, 2026](updates/privacy.md)_

## Overview

This Privacy Policy explains how AIsh.li collects, uses, and safeguards information when you visit our website, use our AI agent runtime services, embed our chat widgets, or access our APIs.

When you use AIsh.li directly as a registered user, AIsh.li acts as the data controller for your account data. When you embed an AIsh.li chat widget on your own website, you act as the data controller for your website visitors, and AIsh.li processes visitor chat interactions as a service provider on your behalf.

## Information We Collect

1. **Account Information**: When you sign in to AIsh.li (via our authentication provider, Clerk), we receive basic profile information such as your email address and authentication identifier.
2. **Conversation, Usage & Session Data**: To provide multi-turn conversational continuity, manage your configured agents, and prevent abuse, we process and store agent configurations, conversation session messages, token counts, request timestamps, client IP/origin metadata, and credit consumption.
3. **Widget Site Domains**: When you whitelist website domains to embed chat widgets, we store the allowed origin URLs to verify authorized embedding requests.

## How We Use Your Data

- To authenticate your access and manage your subscription credits.
- To process and maintain conversations with your configured AI agents and connected tools.
- To enforce domain origin security, rate limits, and abuse prevention for APIs and embeddable widgets.
- To monitor platform performance, diagnose technical errors, analyze aggregated usage trends, and provide customer support.

## AI Model Processing & Zero-Training Commitment

> [!IMPORTANT]
> **Zero Training on Your Data**: Your conversation prompts, inputs, and visitor messages are processed to generate AI responses and maintain your session history via commercial inference APIs (such as Google Gemini). AIsh.li does **not** use your prompts, conversations, or proprietary configurations to train or fine-tune AI models. (When using Bring Your Own Key, your third-party provider's terms apply; see [Free-tier BYOK keys](key.md#free).)

Upstream AI providers process inference requests under commercial data protection terms that prohibit training foundation models on paid API traffic (subject to standard provider safety and abuse monitoring). If you configure your own third-party API keys (BYOK) or external tools/webhooks, requests routed through those credentials are governed directly by your respective agreements with those providers. In particular, while paid commercial API tiers prohibit provider model training, third-party free tiers (such as Google AI Studio's free tier) permit the provider to use inputs and outputs for model training and human review (see [AI Key details](key.md#free)).

## Third-Party Sub-Processors & Infrastructure

We rely on trusted third-party cloud infrastructure partners who act as our data sub-processors under strict contractual data protection agreements:

- **Stripe**: Payment processing and customer billing.
- **Clerk**: Secure identity management and user authentication.
- **Cloudflare**: Global content delivery network (CDN), DDoS mitigation, and edge worker routing.
- **Google Cloud**: Application hosting (Cloud Run), managed database storage (Firestore), and commercial AI inference APIs (Gemini) subject to enterprise zero-training data protection terms.
- **Google Analytics**: Aggregated website traffic, interaction telemetry, and cross-session analytics on our web portal (delivered through first-party edge proxying via Cloudflare Google Tag Gateway).
- **Sentry**: Error monitoring, crash diagnostics, distributed performance tracing, and privacy-masked session replay (hosted in the European Union via `de.sentry.io`).

## Cookies & Analytics (Google Consent Mode v2)

AIsh.li uses browser storage and analytics to operate and improve the service:

- **Essential Local Storage & Authentication**: We use session tokens and browser `localStorage` keys (including a randomly generated `anonymous_id` UUID, UI language locale, and currency preference) strictly to maintain your signed-in state, preserve anonymous widget chat continuity, and remember preferences. We do not use third-party advertising cookies.
- **First-Party Analytics & Google Consent Mode v2**: We configure Google Analytics via Google Consent Mode v2 with first-party analytics storage enabled (`analytics_storage: 'granted'`) and all advertising and personalization storage denied by default (`ad_storage: 'denied'`, `ad_user_data: 'denied'`, `ad_personalization: 'denied'`). First-party analytics cookies (`_ga`) are used strictly to distinguish unique visitors and measure session continuity on our site, never for cross-site advertising or profiling. Measurement scripts and telemetry requests are routed through our own domain (`aish.li/gtag/...`) using Cloudflare's Google Tag Gateway edge proxy rather than directly to third-party endpoints. For signed-in users, we synchronize an opaque, pseudonymous Clerk user identifier (`User-ID`) without any personally identifiable details (no email or name) to measure retention and account usage across sessions. You can block or disable analytics at any time via browser settings or content blockers.
- **Error Monitoring & Privacy-Masked Session Replay (Sentry)**: We use Sentry (`*.ingest.de.sentry.io`, EU region) to diagnose client and server errors, trace request performance, and record diagnostic session replays on our web portal (session replay is disabled inside embedded chat widgets). Sentry does not set any cookies; when a portal session is sampled for replay (`10%` of sessions, or `100%` when an error occurs), it uses a temporary browser `sessionStorage` entry (`sentryReplaySession`) solely to maintain replay continuity within the active browser tab. Session replay runs with strict client-side masking enabled by default (`maskAllText: true`, `maskAllInputs: true`, `blockAllMedia: true`), replacing all on-screen text and user inputs with asterisks (`*`) in your browser before transmission so your prompts, chat messages, and secrets are never captured in replays. For signed-in portal users, telemetry is linked to your pseudonymous Clerk user ID and username (without your email address) to help troubleshoot account-specific issues. You can block client-side error and replay telemetry at any time via browser privacy shields or content blockers.

We do not sell your data to advertisers ourselves.

## Data Retention & Your Rights

We implement technical and organizational security measures to protect your data, though no network transmission or electronic storage can be guaranteed to be 100% secure. Our services are not directed to children under 16.

You have the right to request access to, correction of, or deletion of your personal account data, subject to applicable legal exceptions (such as retaining mandatory billing/tax records, credit ledger audit logs, and security or fraud-prevention identifiers).

If you have questions about this policy or wish to request data deletion, please contact us at <a href="#" data-obfuscated-mailto="bWFpbHRvOnN1cHBvcnRAYWlzaC5saQ==" class="obfuscated-email"><span class="email-user">support</span><span class="email-at" aria-hidden="true"></span><span class="email-domain">aish.li</span></a>.

## Changes to This Privacy Policy {#changes}

We may update this Privacy Policy from time to time to reflect new features, infrastructure sub-processors, operational practices, or regulatory requirements. Any revisions will be reflected in the "Last updated" date at the top of this page, and material changes are recorded in our [Privacy Policy Updates](updates/privacy.md) log. For material changes that expand how we process your personal data or introduce new core data sub-processors, we will provide reasonable advance notice (such as via a prominent notice in the application or by email to the address associated with your registered account) before the updated policy takes effect.
