SearXNG's Creator Ships Hister 0.19, a Search Engine for One

Hister, a self-hosted search engine that indexes the web pages a person actually visits rather than crawling the open web, reached version 0.19.0 on Sept. 3, adding extractors that preserve full ChatGPT conversations and Hacker News comment threads, according to the release notes. The project is built by Adam Tauber, known online as asciimoo, who created the metasearch engine Searx in January 2014 before contributors forked it into SearXNG in 2021 after Tauber stepped back from active maintenance.

What it actually indexes

Hister is not a web crawler by default. A browser extension for Chrome or Firefox saves the full rendered content of every page a person opens, and a separate importer can pull in existing browser history, bookmarks, PDFs, Markdown files and local directories, according to the project's homepage. The result is a full-text index a person can search with field filters, phrase matching, wildcards and negation, licensed under the AGPLv3 and self-hosted on SQLite or Postgres, with no telemetry and no external requests once it's running, per the same page.

That pitch is close to the one Tauber made for Searx a decade earlier: reduce dependence on an external search index by keeping control of the data locally rather than routing it through a company's servers. Hister narrows the target from the entire web to one person's own browsing, files and conversations, which puts it in a different lane than the two other self-hosted tools it gets compared to most often.

Project	What it indexes	License
Hister	Pages a person visits, local files, imported history	AGPLv3
SearXNG	Results aggregated from other search engines	AGPLv3
ArchiveBox	Archived copies of URLs, browser history, bookmarks	MIT

SearXNG, with 37,356 GitHub stars, still searches the open web, just without tracking the person doing the searching. ArchiveBox, with 28,535 stars, preserves a page's full content for later, but is not built around fast day-to-day search the way Hister is.

What 0.19.0 adds

The new ChatGPT extractor keeps user and assistant turns, headings, lists, tables, code blocks and links from both authenticated sessions and public shares, and the Hacker News extractor records a submission's metadata alongside its full nested comment tree, according to the release notes. The release also adds url_re:, a query filter that matches a document's URL against a Go regular expression, and a dry_run mode on the delete endpoint that counts how many documents a proposed deletion rule would remove before anything is actually deleted.

Hister also runs an MCP server, using the Model Context Protocol that lets an AI model call out to external tools over a defined interface, so a coding agent or chat assistant can query a person's own indexed history and files directly rather than relying on whatever the model already knows or a general web search. That puts it alongside a wider move toward running AI agent infrastructure on hardware a team or a person controls rather than a vendor's cloud, the same instinct that shows up in Nix-based tooling built to run entirely inside a browser tab.

What the community flagged

Tauber posted the release to Hacker News, where a commenter using the handle corndoge asked how Hister differs from ArchiveBox, an existing self-hosted web archiving tool. Tauber replied that Hister is built for "creating an active knowledge base and finding information quickly," while ArchiveBox focuses on preserving web content for the long term rather than searching it day to day. Another commenter, g_host, reported that an Nginx reverse proxy placed in front of Hister crashed and stayed unreachable until restarted; Tauber asked for a GitHub issue to diagnose it, and no fix had landed in the thread as of Sept. 20. A third commenter, computator, wrote that installing unpackaged, independently distributed software like Hister carries its own risk regardless of what it promises to do with data once it's running.

Where it doesn't reach

Hister only indexes pages a browser extension actually renders, so it does not follow links out from a page the way a crawler would; a commenter identified as mircea asked whether opening a Hacker News thread also captured the pages it links to, and Tauber confirmed it does not. The project also has no hosted or managed option: every deployment is self-run, which is also why a request in the same thread for one-click hosting through a service called Pikapods went unanswered beyond Tauber naming the missing piece as configuration tooling that does not yet exist.