Sherpa — Local Help Search

Crawl a documentation site, index it entirely in your browser, and answer questions from that index — no servers, no per-query cost.

Sherpa turns any help centre or documentation site into a searchable, question-answering index that lives entirely on your computer. Point it at a docs site. It crawls the pages you can already see, builds a search index in your browser, and answers questions from that index — with a citation on every answer, linked to the exact passage. WHY IT IS DIFFERENT • Nothing leaves your machine. The embedding model and its runtime are bundled inside the extension, so even the first index needs no download. No server, no account, no telemetry, no per-query cost. • It reaches pages other AI tools cannot. Sherpa crawls as you, reusing your existing browser session, so an internal wiki or a customer-gated help centre works the same as a public one. Your credentials are never seen or stored. • Answers are grounded, or it says so. Every answer comes from the pages it indexed. When the index does not cover your question, Sherpa tells you and shows the nearest pages instead of guessing. HOW IT WORKS 1. Open the side panel and choose a site to index. 2. Sherpa shows how many pages are in scope, how long it will take and how much disk it needs — before it fetches anything. 3. It crawls politely: honouring robots.txt and Crawl-delay, one request per second by default, resumable if you close the browser. 4. Ask questions. Answers stream in with numbered citations you can click through to the source passage. FEATURES • Answers with citations, and a "copy answer with citations" button for pasting into a ticket reply • Multiple indexed sites with a switcher, and saved conversation history • Incremental refresh — only changed pages are re-fetched and re-indexed • A content gap report showing which questions your docs failed to answer, exportable as CSV or Markdown • Three answer modes: extractive (no model, works everywhere), Chrome's built-in Gemini Nano, or your own API key • One-click delete of everything Sherpa has stored PRIVACY Sherpa has no backend. The only network traffic is fetching the site you asked it to crawl. There is no analytics, no accounts, and no data collection of any kind. The single exception is opt-in: if you configure your own API key for answering, your question and the retrieved passages go to that provider. The interface says so before you enable it, and it is off by default.