GitHub

GitHub is the one connector whose streams are alternatives rather than additions. A repository is a single file tree, and only you know what it is: a codebase, where one file tells you nothing and the structure is the whole point, or a folder of documents that happen to live in git.

So a repository is read one way or the other, and switching between them is a deliberate act.

Codebase

Pensieve's source-tree agent researches the repository as a whole and maintains a documentation Page tree for it, rather than importing raw code. The repository's root page is its overview; the agent builds whatever nested structure the code actually warrants underneath. Pages in your main tree can link across to repository pages when a project or system depends on that code.

Both readings stay current the same way: every push to the selected branch triggers an update, pinned to the exact commit it saw.

Knowledge

The same repository read as ordinary documents: eligible files are mirrored as individual sources, exactly like Drive or Notion content. Use this for a docs repo, a handbook, or a folder of specifications.

You refine the selection with include/exclude patterns, and the picker previews exactly which files qualify before you save. Document formats are eligible; source code isn't — that's what Codebase reading is for.

Multiple repositories

One GitHub connection can carry many repositories, each with its own branch and its own choice of reading.

Because you added each one, you can disconnect a single repository without touching the connection or the others — and you're asked whether to keep what it contributed or delete it, exactly as you would be for the whole connector. Keep, and it stays in your map and can be deleted later from Data → Sources.

What it can't see

• Repositories you don't connect.
• Anything on a branch you haven't selected.