Skip to content

Nimrod 2.1

Choose a tag to compare

View all tags
@mixelpixx mixelpixx released this 02 Oct 19:33
v2.1.0
36431f8

Nimrod 2.1 reads pages much better: their main content, PDFs included, in sizes an AI can use. Search results now carry dates and related questions, and research picks better sources.

Update

  • Already installed: nimrod update
  • macOS and Linux: curl -fsSL https://nimrod.orchis.ai/install.sh | sh
  • Windows (PowerShell): irm https://nimrod.orchis.ai/install.ps1 | iex
  • Claude Desktop only: open nimrod.mcpb.

Reading pages

  • A page comes back as its main content: the article, docs or post, with menus, ads, cookie banners and sidebars removed. Links are absolute, tables keep their columns, and formulas are read once.
  • PDFs are read: papers, datasheets, reports.
  • Long pages come in parts of 20,000 characters, each ending with where to continue and the page's other sections. A single read can't flood the context window any more.
  • extract_multiple_webpages returns each page's text, for up to 10 pages at once. It used to return a one-line summary per page.
  • Stack Overflow questions (with their top answers) and crates.io crates (with their versions and README) are read through those sites' public APIs.
  • Pages built with JavaScript, bot checks and sign-in walls get a clear message saying what happened and what to try instead.

Searching

  • Each result shows its site and date, with the calendar date next to "2 days ago". Clean URLs, and no more percentage scores.
  • Related questions and related searches, when Google has them.
  • New country filter; date filters from the last hour to the last year. Image results include the image URL.
  • Mistyped filters are refused with an explanation, and don't cost a credit.

Research

  • Sources follow Google's ranking, at most two per site.
  • Pages that can't be read are replaced by the next result.
  • Depth sets how many sources are read (3, 5 or 8), and the searches run in parallel.

New and optional: page scraping (costs credits)

nimrod scrape on (or the "Scrape pages" switch in Claude Desktop) adds scrape_webpage, which renders a page in a full browser remotely, for pages Nimrod can't read on your computer. It costs 1 to 5 credits per page, more than a search, and it's off unless you turn it on. Reading further into a page you've scraped is free.

Also

  • nimrod setup and nimrod status find Claude Code when it's installed only as part of the Claude desktop app.

Builds aren't code-signed yet; the install commands avoid security prompts. Checksums are in SHA256SUMS.txt.

Assets 11
Loading