Outputs
An output is something an agent produced that is worth keeping: a report, a document, a dataset, a diagram, a small app. Instead of being buried in a transcript, it gets its own page.
Outputs come in two kinds.
- File: a result that fits in one file. Lists show it as a Doc, Sheet, Page, or Image.
- App: a result made of several files that reference each other. It opens as a working screen rather than a document.
Both kinds share one list, Outputs in the left rail, newest first. You can find one by name without remembering which kind it is.
When you get one
Section titled “When you get one”The agent decides. Ask for something substantial, such as “write up the findings” or “produce the weekly summary”, and it saves the result as an output and links it in the conversation. You can also ask directly: “save that”. When the result is several files (a page with its styles and scripts), the agent saves it as one app.
Short answers stay in the conversation, which is the right default. An output is for work you will come back to.
Working with them
Section titled “Working with them”In the conversation, an output appears as a card you can open without leaving the thread. The Outputs in this conversation button on the right rail gathers the files and apps that conversation made.
On its own page, you get the full content, a preview for the formats that support one, and export. The agent can read its own outputs back later, which is what makes a chain of work possible: draft in one conversation, revise in the next.
In the agent’s Inventory, the apps and files it has produced are listed. Pin the files you return to so they stay at the top.
Bundling files into an app
Section titled “Bundling files into an app”When an agent saved several files that work together (a page, a small tool, a dashboard), you can bundle them into one app. An app has its own page, opens as a working screen, and can be shared by link the same way a file is.
Bundling is a move: the files become the app’s first version and the original entries go away. That keeps one copy of the truth, so the same result does not appear twice in your lists. The originals are removed only after the app is saved. If saving fails, nothing is lost.
An app opens in one of two modes:
- Viewer: the result, and nothing else around it. This is what a link recipient sees.
- Canvas: the app beside the conversation, so you can ask for a change and watch it land.
Sharing by link
Section titled “Sharing by link”An output is private to you by default. Turning on link sharing produces a URL that opens that one output.
What a link grants:
| Grants | Does not grant |
|---|---|
| Read access to this output | Access to any other output |
| Nothing else | The conversation it came from |
| Your agent, its memory, or your account |
Anyone with the link can open it, so treat it as public. Turning sharing off makes the link stop working. An output that is not shared is not found by URL at all, whether or not someone guessed it.
Older /artifacts/… and /apps/… links still open. The address changes to /outputs/… and the same page appears.
Limits
Section titled “Limits”- Saving a file to the same path again replaces its contents. The earlier state can be restored from the version history on the file’s page.
- Apps cannot be pinned. Pinning works for files only.
- Outputs belong to the agent that produced them and to you. Another agent does not read them unless you point it at one.
- Link sharing is read-only, and it is one link per output. There is no per-person access.
- Deleting an agent does not delete its outputs.