1. Introduction
This privacy policy describes how LIA, an open-source personal AI assistant, collects, uses, and protects your personal data. LIA is developed and operated by an independent developer as an open-source project under the AGPL-3.0 license.
LIA is currently offered free of charge. The application is accessible at https://lia.jeyswork.com. The complete source code is publicly available, allowing you to audit how your data is processed at any time.
This policy applies to the hosted instance of LIA. If you deploy your own instance (self-hosting), you control its operation and must assess the data-protection duties applicable to your use; this policy does not directly apply. We nevertheless encourage you to use it as a starting point for that assessment.
By using LIA, you acknowledge that you have read and understood this policy. If you do not accept the terms described herein, please do not use the service.
2. Data collected
LIA processes the following categories of data to provide the service and the optional features you choose to use:
User account data:
- Email address (unique identifier)
- First and last name
- Password (hashed via bcrypt, never stored in plain text)
- Language preferences and timezone
- User role (standard or administrator)
Conversation data:
- Messages exchanged between you and the assistant
- Execution plans generated by the planning system
- Results of actions performed by agents (email search, event creation, etc.)
- Conversation history, saved as checkpoints in PostgreSQL
- Memories, preferences, documents and other content you provide to personalize the assistant
Third-party service connection data:
- OAuth access and refresh tokens for services that use OAuth, including Google Workspace and Microsoft 365
- Application-specific passwords or other credentials for connectors such as Apple iCloud, and provider API keys you configure
- These stored secrets are encrypted with Fernet (AES-128-CBC with HMAC-SHA256 authentication)
Optional feature data:
- A home address you save, and browser location if you grant permission; remembering the last known position also requires opt-in. These location fields are encrypted, the remembered position replaces the previous one without building a location history, and disabling that option erases it
- Audio, transcripts and any images or documents used in a voice, meeting or multimodal request
- Health measurements if you choose to connect a source and use the health features
Usage data:
- Aggregate operational metrics (request count, response times) and per-account usage records
- LLM token counters consumed per session
- Technical error logs with controls to redact secrets and personal content; optional diagnostic traces have a separate scope described below
Data that LIA does NOT collect:
- Biometric authentication templates held by your device when you use a passkey
- Browsing data outside of the application
- Advertising profiles or targeting data
3. Legal basis for processing
The following table identifies the legal bases used for the hosted service under the General Data Protection Regulation (GDPR):
| Processing activity | Legal basis | Justification |
|---|---|---|
| Account creation and management | Performance of contract (Art. 6.1.b) | Necessary to provide the service |
| Conversations with the assistant | Performance of contract (Art. 6.1.b) | Core function of the service |
| Third-party service connections (Google, Apple, Microsoft) | Explicit consent (Art. 6.1.a) | You actively choose to connect each service |
| Sending data to LLM providers | Performance of contract (Art. 6.1.b) | Necessary for the assistant to function |
| Technical logs and metrics | Legitimate interest (Art. 6.1.f) | Maintaining security and reliability of the service |
| Language preference cookie | Consent (Art. 6.1.a) | Storing your language choice |
You may withdraw your consent at any time for processing based on consent, without affecting the lawfulness of processing carried out prior to withdrawal.
4. Hosting and data location
Hosted instance infrastructure:
The official LIA instance is self-hosted on a physical server administered by the developer. Persistent application data, including accounts, conversations and checkpoints, is stored in France.
- Database: PostgreSQL for persistent storage (accounts, conversations, checkpoints)
- Cache: Redis for sessions and temporary caching
- Reverse proxy: Cloudflare Tunnel for secure HTTPS access
- TLS certificates: Managed automatically by Cloudflare
Public presentation media:
Preview images and videos can load directly from the static media host configured by the operator. On the official instance, the presentation is published in a public GitHub release and delivered through its infrastructure. Loading these files sends the usual technical request information to that host, including the IP address and browser headers. These media show fictional examples, separate from the personal content stored by LIA.
International data transfers:
When you interact with LIA, data needed for your request can be transmitted to model, voice, avatar, search or other service providers, depending on the features and configuration you use. These providers may have servers located outside of the European Union (notably in the United States and China). See the "LLM Providers" section for details.
Connections to Google Workspace, Apple iCloud, and Microsoft 365 also involve exchanges with these providers' servers, subject to their own privacy policies.
5. Data security
LIA implements a multi-layered security architecture designed to protect your data at every stage:
BFF (Backend-for-Frontend) architecture: The application session uses an HttpOnly cookie, and long-lived connector credentials and model or avatar API secrets are managed server-side. Some voice and avatar connections use temporary session credentials in the browser. Interactive Google Maps also receives a browser API key after authenticated activation; that key requires provider-side restrictions and suitable quotas.
Encryption of sensitive data:
- Stored connector credentials, provider secrets and designated location fields are encrypted with Fernet (AES-128-CBC + HMAC-SHA256 authentication)
- Passwords are hashed with bcrypt (adaptive cost factor)
- Access to the hosted application uses HTTPS; protection of internal connections and backups depends on the deployment
This field-level encryption does not encrypt every database column. Conversations are readable by the server for processing and are not end-to-end encrypted. An operator with database and encryption-key access can access stored data; the security of that access and of backups also depends on operation of the instance.
Personal information and diagnostics: Redaction controls protect technical logging; they do not automatically anonymize the context sent to a model. Your messages and relevant connected-service content may contain personal information needed for your request. If LLM tracing is enabled, diagnostic tools can also retain model inputs, outputs and account-linked metadata. Their access, hosting and retention must be configured by the operator.
Sessions and authentication:
- User sessions are stored in Redis with automatic expiration
- Authentication relies on secure cookies (HttpOnly, Secure, SameSite)
- Client-side JavaScript cannot read the HttpOnly application session cookie; temporary credentials for enabled browser voice or avatar sessions have a different purpose
Anti-robot check (optional): When the operator enables it, the sign-up, sign-in and password-reset pages run Cloudflare Turnstile. In a frame served by Cloudflare, it analyses technical signals from your browser to tell a person from a script, then hands the page a single-use token that LIA's server checks with Cloudflare. LIA keeps nothing of these signals and sets no cookie for this check; Cloudflare processes them under its own privacy policy.
Secure logging: Technical logs use structured JSON format (via structlog), with rules to omit quoted user content and redact recognized secrets and personal information. These controls do not make every diagnostic store anonymous.
Commands and private continuations
Keys, file choices, account connections and browser consent use the operation’s secure interface, rather than parameters invented by the model. A command is tied to the account and its authorised context; permissions and the resource version are checked again before writing. Encrypted command content does not hide the whole service from its operator: messages and operational data retain the limits described here.
6. LLM providers
LIA uses multiple Large Language Model (LLM) providers to process your requests. The choice of provider depends on your instance configuration and the type of task:
| Provider | Headquarters | Usage in LIA |
|---|---|---|
| OpenAI | United States | GPT models for conversation and planning |
| Anthropic | United States | Claude models for conversation and analysis |
| Google (Gemini) | United States | Gemini models for multimodal processing |
| DeepSeek | China | Advanced reasoning models |
| Qwen (Alibaba) | China | Language processing models |
| Perplexity | United States | Augmented web search |
| Ollama | Depends on the configured endpoint | Model inference on a local server when configured that way |
What is transmitted to LLM providers:
- The content of your messages
- Conversational context necessary for response coherence
- Relevant tool results (email content, event details, documents, etc.), which can contain personal data
Voice, meeting, image and optional speaking-avatar paths can additionally transmit the audio, text or images needed by their respective providers. The selected features and routing determine which services receive them.
How credentials are used: Connector credentials and provider keys are used to authenticate to the relevant service, not added to model prompts as conversation content. Avoid putting passwords or other secrets in your messages: content you supply can become part of a request to a provider.
Provider commitments: Training use, retention and other handling of data depend on the provider, product, account settings and applicable contract or privacy policy. Review those terms for each service you enable; LIA cannot make a universal no-training guarantee on their behalf.
Choosing a locally hosted Ollama endpoint keeps inference for the selected model on that endpoint. It does not make other integrations local: connected accounts, web search, remote voice or avatar services can still exchange data with their providers.
7. Data retention
Retention periods are defined based on the nature of the data:
| Data type | Retention period | Justification |
|---|---|---|
| User account | Personal content is purged at the deletion step; the account row and billing records remain until their applicable erasure or retention step | Service operation, accounting and applicable legal duties |
| Conversation history | Until deletion by user or account deletion | Service continuity |
| Encrypted connector credentials | Until service disconnection or account deletion | Access to connected services |
| Remembered browser location | Replaced on update and erased when its opt-in is disabled; freshness limits its use | Location-based requests without a location history |
| Redis sessions | Automatic expiration according to session and remember-me settings | Security |
| Technical logs and optional diagnostic traces | Retention configured for each diagnostic store | Diagnostics and security |
| Usage metrics | Retention configured for the metrics store; account-linked records follow their documented lifecycle | Usage accounting and service operation |
Account deletion: You can request account deletion from the administrator. The deletion step purges personal content such as conversations, memories, documents, checkpoints and stored connector credentials, but retains the account row, including name and email, and billing records. The subsequent erasure step removes the remaining account row. Audit records, separate diagnostic stores, backups and data already received by providers require their own retention and erasure procedures; neither step instantly erases every copy. The operator must handle applicable erasure requests across those stores.
8. Your rights
Under the GDPR, you have the following rights:
- Right of access (Art. 15): Obtain a copy of all personal data we hold about you.
- Right to rectification (Art. 16): Correct inaccurate or incomplete personal data.
- Right to erasure (Art. 17): Request the deletion of your personal data ("right to be forgotten").
- Right to restriction (Art. 18): Request restriction of processing of your data in certain circumstances.
- Right to data portability (Art. 20): Receive your data in a structured, commonly used, and machine-readable format.
- Right to object (Art. 21): Object to the processing of your data based on legitimate interest.
- Right to withdraw consent: At any time, without affecting the lawfulness of processing carried out prior to withdrawal.
To exercise these rights, contact us at the address indicated in the Contact section. We will respond without undue delay and normally within one month of receiving your request, with any justified extension communicated as provided by Article 12 of the GDPR.
If you believe your rights are not being respected, you have the right to lodge a complaint with the CNIL (Commission Nationale de l'Informatique et des Libertes) or any other competent supervisory authority.
9. Cookies
LIA uses a minimal number of cookies, exclusively functional:
| Cookie | Purpose | Duration | Type |
|---|---|---|---|
NEXT_LOCALE | Stores your language preference (fr, en, de, es, it, zh) | 1 year | Functional |
| Session cookie | Maintains your authentication session | Session duration | Strictly necessary |
What LIA does NOT use:
- No tracking cookies
- No advertising cookies
- No third-party analytics cookies (Google Analytics, etc.)
- No tracking pixels
- No browser fingerprinting by LIA (the optional anti-robot check of section 5 is Cloudflare's)
The cookies used by LIA are either strictly necessary for the operation of the service or relate to your explicit choice (language preference). In accordance with the ePrivacy Directive, strictly necessary cookies do not require prior consent.
10. Contact
For any questions regarding the protection of your personal data, the exercise of your rights, or this policy, you may contact us:
- Email: [email protected]
- Website: https://lia.jeyswork.com
- Source code: GitHub (AGPL-3.0)
Data controller: LIA is operated by an independent developer acting as data controller within the meaning of the GDPR.
Changes to this policy: This policy may be updated to reflect changes in the service or applicable regulations. In the event of a substantial modification, you will be notified through the application. The update date at the top of this document prevails. We encourage you to review this policy regularly.
Open-source transparency: As an open-source project, LIA allows you to audit the source code at any time to verify exactly what data is collected, how it is processed, and where it is sent. This radical transparency is a fundamental commitment of the project.