Praex: Privacy Policy

Effective 2026-10-22 (first version 2026-08-06) · Praex is operated by Lucius AI Systems, Inc (contact: davor@praex.ai)

What Praex is

Praex is an AI coding agent and copartner: a web app, a desktop app, an Android app, and a terminal CLI. It can run against our hosted models (Velox, Faber, Lucia) or against endpoints you configure yourself (bring-your-own-key).

Hosted models: what we process and store

• Your conversations are processed transiently to generate the reply. We do not train models on your content, and we do not retain prompt or reply bodies by default.
• Attachments: a file you attach is read on your device, and its name and text go to our servers inside your message so the model can answer. They are processed like the rest of the conversation and not kept.
• Projects and memory (web chat): a project's instructions and files, and the facts you save to memory, are sent with your messages so the model can use them. They are processed like the rest of the conversation. Memory holds only what you save yourself; nothing is added to it for you.
• Reply reports: if you report a reply, we keep the reason you picked, your optional note, the reported reply text (up to 2,000 characters), the model name and your account ID, so we can review it. If our servers can't be reached, the app opens an email to davor@praex.ai with the same text instead.
• Stand-in models: while our own Faber and Lucia servers are not running, those tiers are answered by third-party open models (currently DeepSeek-V4.1-Flash for Faber and Kimi K2 Thinking for Lucia) served through Google Cloud Vertex AI. For those replies your conversation is processed by Google Cloud under its Vertex AI terms. Velox runs on our own servers.
• Web search: when a model searches the web for you, the search query (not your conversation) is sent to a web search provider.
• Usage metering: we store daily token counts keyed to your account ID, and your subscription plan, to enforce allowances and billing.
• Guest use: if you chat on the web without signing in, an anonymous account is created when you send your first message, so the guest allowance can be counted. Guest-trial limits are also counted per network: we keep a hashed form of your IP address (not the address itself) in our usage records, with that day's token count.
• Your chat history stays on your device (browser storage, the app, or the CLI's local database). We do not keep a copy, except for conversations you choose to share, and chats, projects and memory you choose to sync to your account (Settings, off by default). Synced data is deleted when you turn sync off and delete it, or delete your account; deleted data can remain in our storage backups for up to 7 days.
• Sign-in: Google sign-in via Firebase / Google Identity Platform. We never see or store passwords.
• Billing: payments are handled by Stripe; we store your Stripe customer and subscription IDs, never card numbers.
• Android subscriptions are billed through Google Play. To verify the subscription we store the Play purchase token, its status and expiry date, and a mapping from a one-way hash of your account ID to your account. Payment details stay with Google.

Sharing, sync and Remote Control (only when you turn them on)

• Chat sync: when you turn on "Sync chats to your account" in the web chat's Settings (off by default, Google accounts only), your chats, projects and memory are stored on our servers (Google Cloud Storage in the EU) under your account, so they open on your other devices. Turn it off to stop syncing, then delete them from your account in the same place, or delete your account. Chats stay in your browser either way.
• Share: when you share a conversation (/share or the Share button), a read-only copy of its messages is stored on our servers and anyone with the link can read it. Tool output and file contents are left out. It stays until you delete it (/unshare) or delete your account. Links made before 20 October 2026 are not removed with the account; remove them with /unshare or email us the links.
• Remote Control: when you turn on /rc, traffic between your computer and your phone or browser passes through our relay server so you can continue the session there. The relay does not store your conversation; it only keeps track of which sessions are connected while they are connected.

Bring-your-own-key mode

• Conversations go to the model endpoint you configure (your own server or a third-party provider under their terms) and are not sent to us.
• Settings, endpoint URLs, and credentials you enter are stored locally on your device.

Service providers we rely on

Google Cloud, in US and EU regions (hosting, sign-in, usage metering, and Vertex AI for the stand-in models), Stripe (payments), Google Play (purchases inside the Android app), and web search providers (for search queries only). They process data only to provide their service to us.

What we do not do

• No advertising, no trackers, no analytics SDKs, no sale or sharing of personal data.
• No collection of contacts, location, or device identifiers.
• No training on user content, hosted or otherwise.

Retention, access, deletion

Usage counters and billing records are kept while your account exists and as required for accounting. Shared conversations are kept until you delete them or your account (links made before 20 October 2026: see Sharing above). Synced chats, projects and memory are kept until you delete them in Settings or delete your account; deleted data can remain in our storage backups for up to 7 days. You can delete your account yourself in Settings (see praex.ai/account/delete). Email davor@praex.ai to ask what we hold about you or to have your account and its data deleted; we honor such requests within 30 days. EU/EEA users: you also have the rights of access, rectification, portability, and complaint to your local supervisory authority.