GenizahSearch AI integrations — Privacy Policy

Last updated: 10 October 2026

This policy covers GenizahSearch's AI integrations: the MCP server at https://genizahsearch.com/mcp, which ChatGPT, Claude, Gemini and other MCP clients connect to, and, until OpenAI retires custom GPTs, the actions used by the GenizahSearch GPT in ChatGPT. The integrations are maintained by Hillel Gershuni, creator of Dicta Genizah Search, with the support of Dicta. Contact: [email protected]. See also the terms of use and the support page.

Information received

When an AI assistant calls a GenizahSearch tool or action, it sends GenizahSearch the parameters needed for that request: search phrases, passages or witnesses, spelling and matching options, filters, manuscript/page identifiers, and job identifiers used to retrieve results. This may include text taken from your messages or uploaded files. GenizahSearch does not request your name, email, password, or full conversation. Only content included in tool or action requests is sent to GenizahSearch. An MCP client may also send the MCP protocol version and the name it gives itself (for example, the name of the AI application).

Our infrastructure also receives technical request information, including connection IP addresses (often those of the service making the call), headers, request paths, timestamps, and response status. Job identifiers may appear in request paths.

How information is used

We process requests to search the manuscript corpus, retrieve pages, find parallels, return results to the AI assistant that called the tool, limit server load, prevent abuse, and diagnose failures. The integrations do not create researcher accounts or publish submitted passages. They do not use submitted research text to train an AI model or serve advertisements.

Temporary storage and operational records

The MCP server keeps no conversation state between calls. Research jobs hold request information and results in server memory. Completed jobs become unavailable ten minutes after completion; expired entries are removed on subsequent job requests or server shutdown. Restarting the service clears this memory. Search workers also use temporary files for inputs and outputs, normally removed when worker processing ends; an abnormal server termination can leave temporary files behind.

Operational logs and configured service analytics may last longer than the job cache. Logs may contain technical request information and error diagnostics. Server logs, meaning the application's log files and the service's console log, can also include search queries, search patterns derived from them, and error messages that contain request content; they are used to diagnose failures and are kept according to the server's log settings, not the ten-minute job expiry. When enabled, PostHog receives limited API usage metrics such as endpoint, search mode, duration, result count, status, and a hashed client-IP identifier; the API metrics event does not include submitted search text. For MCP tool calls, PostHog also receives the tool name, its outcome and any error code, a duration range, the MCP protocol version, the name the client gives itself when it sends one, and the same hashed client-IP identifier; these events never include tool arguments, search text, or results. Analytics are best-effort and may be sampled. Operational records follow the hosting and analytics systems' retention settings, rather than the ten-minute result expiry.

Other services and access

Results are returned to the AI assistant that called the tool. Hosting providers and, when configured, PostHog process operational data to support the service. Administrators may access records for maintenance, security, and support. Anyone possessing an unexpired job identifier can retrieve its results; keep these identifiers private.

The provider of your AI assistant separately processes your conversations, uploads, tool calls, and generated files under its own privacy policy and your account settings: for example OpenAI, Anthropic, or Google. GenizahSearch's job expiry does not delete copies held by that provider. Following manuscript or image links can send requests to holding libraries and other providers, whose own privacy policies apply.

Your choices and requests

You can use the website directly or stop using the AI integrations. Send only the text needed for research; avoid submitting confidential material or unnecessary personal information. For questions or requests to access, correct, or delete personal information held by GenizahSearch, email the contact above. Provide an approximate request time and enough context to locate a record, without posting private job identifiers publicly. We may need to verify a request and may be unable to recover expired data. Contact your AI provider separately about information it holds.

Changes

Updates will be published on this page with a revised date.