How Contextory handles your data

This page describes the product as it works today. Contextory is a hosted service, not a local-first, end-to-end encrypted, or zero-knowledge product. Our systems and the providers below process readable content to provide imports, memory extraction, summaries, and retrieval.

Last updated: September 19, 2026

Data we handle

Account data: your email address, Supabase authentication records, and session information.
Conversation data: conversations and messages you import or sync, plus attachment names, metadata, and extracted text.
Derived data: retrieval chunks, extracted memories, weekly summaries, and vector embeddings generated from your content.
Activity data: import and sync job history, retrieval queries and timestamps, MCP token metadata, and MCP tool activity. MCP access tokens are stored only as hashes.
Waitlist data: an email submitted to the waitlist is stored separately from an account.

We use this data to authenticate you, import and sync conversations, extract and retrieve context, operate MCP connections, show product history, and troubleshoot the service. The current web application does not include an advertising or product analytics SDK.

Where data is stored and processed

Supabase: Authentication and the primary database for account, conversation, memory, job, and activity records.
Pinecone: Vector embeddings and associated metadata used for semantic retrieval. Contextory's tenant namespaces are shared and access is filtered by user ID.
OpenAI: Text is sent to the embeddings API to create vectors for chunks, memories, summaries, and retrieval queries.
Anthropic: Conversation text is sent to Claude for memory extraction and summary generation when the integration is configured; the service has a local heuristic fallback.
Railway: Hosts the API and background worker. Normalized copies of uploaded export archives may also be held on a Railway container's local disk for background import processing.
Vercel: Hosts and delivers the Contextory web application.

These providers receive only the data needed for their role, but their processing is not “zero retention” by a technical guarantee in Contextory. Their handling is also governed by the terms and configurations of Contextory's provider accounts. Contextory does not send your archive back to ChatGPT or Claude as an account-level import; the OpenAI and Anthropic API uses described above are separate processing.

Browser extension and sync

The Chrome extension reads supported ChatGPT and Claude pages after you install it. Captured messages and your Contextory session are kept in Chrome's local extension storage, then pending messages are sent to Contextory when sync runs. Synced captures remain locally until you remove them. Extension preferences are stored withchrome.storage.sync, so Chrome may synchronize them through your browser account; message content and the Contextory session use local storage instead.

An optional blocklist can prevent matching pages or content from being captured and can remove matching pending captures before sync. It cannot retract content that was already sent to Contextory.

Contextory's use of Chrome extension user data complies with the Chrome Web Store User Data Policy, including its Limited Use requirements. Extension data is used only to provide or improve the disclosed context capture, sync, and retrieval features and their related operation and security. We do not sell this data or use it for advertising, creditworthiness, or lending. Transfers to service providers are limited to what is needed for these features or other uses permitted by the policy. Human access is limited to the policy's permitted cases, such as your explicit consent for specific data, security investigations, or legal requirements.

Export, deletion, and retention

You can download a JSON account export from Data settings. It includes stored account content and activity but omits embedding values, uploaded archive files, authentication records, and MCP token hashes.
Deleting a memory in the product marks its database record as deprecated and removes its current vector. The underlying source conversation remains unless separately removed.
Full account deletion is currently unavailable. The deletion control is disabled because Contextory cannot yet verify removal of every historical vector and every upload copy across container-local storage. We do not present a deletion request as completed while those gaps remain.
No automatic retention period is currently implemented for account records, synced captures, retrieval logs, or normalized upload copies. They may remain for the life of the account unless an available product control removes them.

For help with an export or a privacy question, email support@contextory.me.

Security and important limits

Contextory requires authentication for account data and applies user-scoped database queries and row-level security. Service credentials and MCP token verifiers are not included in browser bundles or account exports. These controls reduce risk, but no hosted service can promise absolute security.

Most importantly, content must be readable by Contextory's backend to perform its current features. Do not use Contextory for content that requires client-only keys, end-to-end encryption, or a guarantee that a processing provider never retains data.