Overview
How Silkra works, how to get it on the machine, and where to start.
Silkra is a desktop crawl workspace. You scrape a site on this machine, then work from the table, the semantic map, and an agent that already has that crawl. Nothing is crawled until you confirm the scope.
Download the Mac or Windows app, then sign in. A free account is enough to start. Download Silkra →
Start hereLearn Silkra workflows
A first session
Paste a domain, a subfolder, or a sitemap, preview the scope, and scrape. That crawl becomes a workspace on this machine.
A workspace is one site, or one slice of a site. Each crawl is a version. Refreshing writes the next version in the same workspace. A different site belongs in a new one.
Where you work
The table lists every URL in the crawl. The command bar narrows it by a field, a plain-language question, or by meaning.
The map groups pages that belong together and shows how they link.
The agent answers from that crawl. You can attach the pages the next message is about.
After the crawl
Connections add Search Console, a CMS, or a project tool so those records sit next to the scrape.
Issues are the checks on the crawl, each with a severity and a short explanation.
When you sync a workspace, MCP can expose it to Claude, ChatGPT, Codex, or Cursor.
On this machine
Pages, embeddings, and saved workspaces stay on disk. Group names and the agent use AI through Silkra's gateway, which receives only titles, short excerpts, and the context the agent looks up. The first intelligence run may download an on-device embedding model, about 35MB, once.
Guides for the table, the map, connections, issues, and MCP are in the sidebar.