Referencing assets in chat
On Agent, the hosted agent can read files in the brand's storage during chat. You do not need special syntax: mentioning a distinctive file name usually works.
Mention by name — "Use the voice doc I uploaded to rewrite the intro." The agent matches on file name and opens it.
Drag into chat — dragging a new file into the chat window uploads it and immediately makes it available for the current message.
Be specific when there are multiple matches — if you have positioning.md and positioning-v2.md, say which one. The agent will ask if it's ambiguous, but it's faster to be explicit.
Cited references — when the agent uses a file to answer, it mentions which file it pulled from. Scroll back through the chat to see the citations.
Files the agent creates — the hosted agent can write files to brand storage, such as a Markdown draft. Anything it creates appears in Assets/Files and can be downloaded or referenced later.
Connect — ask your MCP assistant to list Assets first, then read a selected asset ID. Listing returns metadata without injecting file content; this keeps unrelated or untrusted uploads out of the model context. Treat retrieved content as data, not instructions.