Privacy Policy
VEIRA · Last updated August 21, 2026
Overview
VEIRA is a private, general-purpose AI platform for a single owner — there is no public registration and no multi-user accounts. Documents the owner uploads are one of several sources it can draw on, alongside live web search, connected integrations and generated work artifacts. This policy explains what data VEIRA handles, where it is stored, which third-party services it is shared with, and why.
What is stored, and where
Uploaded documents (full content): stored in a private Vercel Blob store and a private Neon Postgres database that belong to the owner's own cloud project. Full document content is never sent to Anthropic or Voyage AI in full — only the fragments described below.
Chat history and metadata: categories, document metadata and chat messages are stored in the same private Neon Postgres database.
Third-party processors
Voyage AI — when a document is uploaded and indexed, and again for every search query, text fragments (typically a few hundred words) are sent to the Voyage AI API to generate vector embeddings used for semantic search. See voyageai.com for their own privacy terms.
Anthropic (Claude API) — fragments the assistant retrieves as relevant to a question (via document search or reading a full document) are sent to the Claude API to generate an answer. If the assistant uses live web search or fetches a web page, the relevant part of the query is sent to Anthropic to perform that lookup.
Google, GitHub, Linear, Notion APIs — only used when the owner has explicitly connected that provider under Integrations, and only to serve the owner's own request in chat. See the scope tables below for exactly what each connection can access.
Google API scopes and data use
The table below lists every Google OAuth scope VEIRA can request, and exactly why. All data scopes are read-only.
| Scope | Why it's requested |
|---|---|
| Basic identity | Confirms which Google account was connected, so the owner can see and disconnect it in Integrations. |
| Gmail — read only | Lets the assistant search and read relevant emails when the owner asks about them in chat (e.g. "find the email from the Belastingdienst about my objection"). VEIRA never sends, replies to, labels, or deletes email. |
| Drive — read only | Lets the assistant find and read the content of Drive files the owner references in chat, so it can be used alongside uploaded documents. VEIRA never creates, edits, moves, or deletes Drive files. |
| Calendar — read only | Lets the assistant look up upcoming appointments or deadlines the owner asks about (e.g. checking a hearing date against a calendar entry). VEIRA never creates or modifies calendar events. |
| Sheets — read only | Lets the assistant read specific cells or ranges from a Google Sheet the owner references in chat. VEIRA never writes to or modifies a spreadsheet. |
How connected-account tokens are protected
Access and refresh tokens for Google, Linear and Notion are never stored in plain text. They are encrypted at rest (AES-256-GCM) in the owner's Neon database, with the encryption key held only in server environment variables — never in the database itself or any database backup. GitHub is connected via a GitHub App instead of a stored user token: only a non-secret installation reference is kept, and short-lived API tokens are minted on demand server-side. The owner can disconnect any provider at any time from the Integrations page, which deletes its stored tokens/installation reference immediately.
Other connected services (GitHub, Linear, Notion)
| Provider | Scopes | Why it's requested |
|---|---|---|
| GitHub | GitHub App installation (owner selects which repositories, from one to all) | Search and read the owner's selected repositories, and — only when the owner explicitly asks — create a new (non-destructive) issue. Access is scoped to exactly the repositories chosen during installation (managed entirely on GitHub's own settings pages, not by VEIRA); nothing is ever closed, deleted, or modified beyond creating a new issue. |
| Linear | read, write | Search and read issues/teams, and — only when the owner explicitly asks — create a new issue. Nothing is ever closed or deleted through VEIRA. |
| Notion | n/a (Notion has no scope parameter) | Search and read only the pages/databases the owner has explicitly shared with the VEIRA integration inside Notion itself. Nothing is ever created, edited, or deleted. |
A custom MCP connector can also be installed by the owner for services without an official integration; VEIRA only sees what that connector's tools expose, and any connector can be disabled or removed at any time.
Memory: what VEIRA remembers about you
Alongside documents and chat history, VEIRA automatically extracts and stores a separate set of memories — preferences, facts, patterns and goals inferred from conversations (e.g. how you like answers formatted, or a recurring fact you've mentioned) — so future conversations can draw on them without you repeating yourself. This happens automatically after substantive replies, not only when you explicitly ask VEIRA to remember something.
Storage: memories are stored in the same private Neon Postgres database as everything else, never with a third party. When a memory is relevant to what you're asking, its content is included in the context sent to the Claude API to generate a response — the same data flow as a document fragment or search result.
Your control: every memory can be viewed, edited or deleted from within the app. A memory can be marked as more sensitive, but — to be precise rather than reassuring — that marking is not yet enforced anywhere in how memories are selected or included in context; it is recorded but currently has no behavioral effect. Deleting a memory removes it from the database immediately.
Data retention and deletion
Uploaded documents, their extracted text and search embeddings are kept until the owner deletes them from within the app, which removes the file from Blob storage and its data from the database. Disconnecting an integration deletes its stored tokens immediately. There is no automated backup beyond what Neon and Vercel Blob provide as part of their own infrastructure.
Contact
VEIRA is operated by a single owner for personal use. Questions about this policy or the data it describes can be sent to Pro.jokhoe2@gmail.com.