Privacy — Search Fragments

Last updated: 1 October 2026

Search Fragments is a free, non-commercial, early-stage prototype run from the UK. It helps AI assistants identify things people only half-remember, and check short factual claims. It works through an MCP server at https://searchfragments.com/api/mcp and a demo on this website. This page explains, in plain terms, what we store and why.

The short version

  • No account needed, no keys. We don’t ask for your name or email.
  • We store the text of queries, the answers we give, any feedback, and a one-way hash of your IP address.
  • Raw queries, logs and identifiers are deleted within 90 days.
  • Solved answers (a title, source links and a short public excerpt) are kept, with the query and any identifiers removed.
  • No tracking cookies: the website uses only Vercel’s cookieless analytics.
  • We never sell data and don’t use it for advertising.

What the service does

There are three tools:

  • resolve_fragment takes a half-remembered description and returns an answer with a source, a shortlist, or “not found”.
  • submit_resolution_feedback records whether an answer was right, with an optional correction, extra clue or short note.
  • verify_claim checks a short factual claim against web sources and returns a verdict with citations.

What we collect

  • Queries and claims: the text you or your assistant sends, and the result we return.
  • Feedback: right/wrong/partly right, plus any correction, clue or note you add.
  • A hashed IP address: a one-way hash of your IP address. We don’t store the address itself.
  • A hashed connection ID: when an AI app connects, it gets a random session ID. We store a hash of it so feedback can be matched to the right answer.
  • Basic request details: time, which app or platform the request came from (for example Claude or ChatGPT, from the user-agent), referring page, approximate country (from our host’s location header), and response time.
  • Website analytics: this website uses Vercel Web Analytics to count page visits. It uses no cookies and doesn’t track you across sites. We don’t use any tracking or advertising cookies.

Please don’t put personal details about yourself or anyone else into queries or feedback. We don’t need them.

Why we use it

  • To answer the query or check the claim.
  • To apply fair-use limits and stop abuse (the IP hash).
  • To match feedback to the answer it is about (the connection hash).
  • To keep the service running, fix problems, and improve answers.
  • To count usage in aggregate (for example, calls per day by platform).

Under UK GDPR our lawful basis is legitimate interests: running a free service, keeping it secure, and improving it. We keep only what those purposes need.

How long we keep it

  • Raw query and claim text, request logs, IP hashes, connection hashes, user-agent and referrer, and feedback notes: deleted within 90 days. A daily job deletes records once they are more than 90 days old.
  • Solved answers: kept indefinitely so the same question can be answered better. We keep only the answer title, source links and a short excerpt from the public source. The original query text and any hashes are removed.
  • Aggregate counts with no identifiers or query text: kept.
  • Backups: deleted data can remain in our database provider’s routine backups for a short time, until they roll over.

Who we share it with

We don’t sell or share data for marketing. To run the service, query text (or parts of it) goes to these providers, who process it for us:

  • Vercel: hosting, cookieless website analytics, and its AI Gateway, which routes some model calls.
  • Supabase: our database, hosted in Ireland (EU).
  • Tavily: web search, to find sources.
  • AI model providers: Anthropic (Claude), and, through Vercel’s AI Gateway, open-model hosts such as Cerebras or Groq. They read the query to understand it and check sources.
  • MusicBrainz and Mixcloud: public music catalogues, searched with terms taken from music-related queries.
  • Resend: sends us occasional internal alert emails, which can include an example query.

Some of these providers are in the US, so data may be processed outside the UK. They each have their own privacy terms. We may also disclose data if the law requires it.

If you use Search Fragments through an AI app (such as Claude or ChatGPT), that app’s own privacy policy covers your conversation with it. We only see what the app sends to our tools.

Your rights

Under UK GDPR you can ask for a copy of your data, or ask us to correct or delete it, or to stop using it. Email hello@searchfragments.com. We don’t know who you are, so it helps to tell us roughly when you sent the query and what it said, or give us the resolution_id. We’ll reply within one month. If you’re unhappy with how we handle your data, you can complain to the Information Commissioner’s Office (ico.org.uk).

Children

The service isn’t aimed at children under 13, and we don’t knowingly collect their data.

Changes

This is a prototype, so this page may change. We’ll update the date at the top when it does.

Contact

Search Fragments, hello@searchfragments.com. This page is governed by the law of England and Wales.

← Search Fragments