silkra

Workspaces

A workspace is one site, or one slice of a site. Not one crawl.

Summarize

A workspace holds the crawls, chat, and artifacts for one job. Most of the time that is one client site. Sometimes it is a slice, like /blog or /en, so the table stays small enough to work in.

A crawl inside a workspace is a version. Switching from v1 to v2 in the title bar does not delete the other run. Refreshing the same site writes the next version in this workspace. A different site, or a totally different scope, belongs in a new workspace.

Create, switch, rename

Starting a crawl creates the workspace. Switch from the rail on the start screen, the sidebar menu, or ⌘P. That palette also jumps to a thread or an artifact in the current workspace, not just other workspaces.

Rename from Manage workspace. That dialog is also where you delete.

Delete and sign out

Delete removes the local crawls, pages, artifacts, chat, embeddings, and intelligence for that workspace. There is no undo.

Sign out does not delete anything. The files stay on this device until you delete the workspace.

If a crawl dies mid-flight, the table banner lets you resume, keep the partial version, or delete it.

Where the data lives

Crawls stay in the local Silkra database on this device. After a crawl, page titles and a few short excerpts go through Silkra's gateway to name the semantic groups; when you use the agent, selected context goes through the gateway to generate the reply. The first intelligence run may download the on-device embedding model, about 35MB.

On paid plans you can attach a second workspace in the agent and compare two crawls in one thread.

How to start a crawl → Refreshing a crawl → Agents →