Skip to Oxygen content

Org Corpus vs Workspaces: Two Ways to Bring Files into a Cloud AI Studio

Serious cloud AI needs two modes: durable institutional memory, and “open this project folder and work.” Oxygen supports both — sticky org corpus uploads into your workspace, and cloud workspaces that load a project into the hosted sandbox.

Org corpus: sticky memory for the team ritual

Upload CSVs, PDFs, and docs into organization-scoped storage. Files land in your cloud workspace and are parsed into advanced storage — including graph and vector layers — so chat, Studio agents, and flows can query the same memory later. This is for handbooks, recurring MLS pulls, board packs, SOP libraries: things you want to keep around.

Cloud workspaces: project folder in the sandbox

When the job is a live project — a repo, a research folder, a client packet — open that folder as a workspace. Oxygen loads it into a hosted sandbox so agents can read and write against the tree. It feels like a lightweight built-in IDE path: browse files on the left, chat in the center, open editors when you need to inspect or tweak.

  • Best for active analysis folders and small builds.
  • Keeps the source of truth in the workspace you opened.
  • Pairs with sandbox packages and the script runner for exploratory work.

Which mode should you start with?

  1. Recurring role playbook (CMA, Monday metrics, policy Q&A) → start with org corpus uploads.
  2. One-off deep dive on a folder you already have → open it as a workspace.
  3. Many teams use both: sticky reference docs in corpus, active exports in a week folder.

Privacy posture stays the same

Both modes live in your cloud workspace. Optional hosted models receive prompt/context you send — not a silent sync of every file to a public chat product.

FAQ

Are org uploads stored in the cloud?
Yes. Org corpus uploads live in your Oxygen cloud workspace — not a public chat history and not a generic consumer drive.
Can agents edit my workspace?
Cloud workspaces are bound into the hosted sandbox so agents can work against that tree. Treat it like granting a tool access to a project folder — open only what you intend to work on.