Workspace

The Workspace page works over the local repository catalog instead of a single repository. It has three tabs: Overview, Activity and Code.

The catalog

The catalog is a list of local Git repositories discovered under roots you add. To build it:

1. Add one or more root directories.
2. Scan. The scan walks each root and records every directory that contains a Git repository.
3. Repositories can be hidden from the catalog without deleting anything from disk, and unhidden later.

The scan is bounded so a home directory with a large tree cannot run forever:

A scan that hits a limit says so in the status line. The catalog stores paths and identifiers only; it does not copy repository data.

Opening a repository from the workspace switches to the workbench with that repository as a tab. To give it its own window, use File → New Window with Repository….

Overview

One row per repository, with:

• Current branch and upstream.
• Staged, changed and untracked counts.
• Ahead and behind counts against the upstream.
• Last commit subject and relative time.
• Operation state (merging, rebasing, cherry-picking, reverting) and stash count.

Search filters by repository name or path. Sorting is by recent activity, name or dirty state. Refresh rereads the status of every visible repository; a repository that cannot be read is reported per row instead of failing the whole refresh.

Select several repositories to fetch or pull them in one action. Each repository gets its own result: succeeded, failed with the Git error, or skipped with a reason (dirty working tree, diverged from upstream, no upstream, unborn branch, detached HEAD, bare repository). Pull is deliberately conservative: it will not merge into a dirty work tree behind your back.

Activity

The activity feed merges recent commits across the catalog, grouped by date.

Filters:

• Free text over commit messages.
• author: for an author string.
• A commit hash or prefix.
• repo: for a repository name.

The feed is capped at 300 entries; the footer says when the cap was reached. Clicking an entry opens the repository at that commit. A refresh that cannot read some repositories reports how many failed and keeps the entries it did read.

Code

The Code tab runs a bounded git grep across the catalog:

• Minimum query length is two characters.
• Fixed-string matching by default, with a regex toggle.
• Case sensitivity toggle and an optional path filter.
• At most 20 matches per repository and 300 matches in total; the cap note states both numbers.
• Binary files are skipped. The search runs against tracked content, as git grep does.

Clicking a match opens the repository and reveals the file and line in the Files preview.

The search is intentionally bounded rather than exhaustive. Narrow the query or add a path filter if you hit the cap.

Windows

• Each window owns its own repository session: open tabs, the current repository, the AI client and the forge client.
• Settings, the local database, the catalog, diagnostics, updates and the MCP server are shared.
• Cmd/Ctrl + N opens a new window. File → New Window with Repository… picks a repository directly.
• The Window menu lists open windows and focuses them.
• At most eight windows exist at once. Closing a secondary window destroys it and releases its Git and AI services.

Next

• Sync — fetch, pull, push and the push preflight.
• Settings — catalog and window behaviour.
• Recovery — snapshots are per repository, not per window.