When you ask an agent to produce something, like a filled-in spreadsheet, a drafted Word document, a response summary, or a populated RFQ, the agent creates an artifact: a file it writes while working on your request. Artifacts appear inline in the chat so you can open them, preview their contents, and download the result. They’re how the agent delivers work that doesn’t fit into a chat message.
Recognizing an artifact in chat
When the agent finishes producing a file, it posts an artifact card directly in the response. The card shows:
- A file icon that reflects the artifact’s type (spreadsheet, document, PDF, image, and so on).
- A title, which is a short label the agent picked, such as “Acme crossover response” or “Q2 summary memo.”
- A subtitle, which is the filename when there is one file (“acme_crossover_2026-04-24.xlsx”) or a file count when the agent produced several (“3 files”).
The card is interactive. Click anywhere on it to open the artifact.
Opening an artifact
Clicking the card opens the artifact drawer on the right side of the chat. What you see inside depends on how many files the artifact contains:
- Single file: the drawer shows an inline preview of the file. You can scroll through pages of a PDF, see the rows of a spreadsheet, read the text of a Word document, or view an image.
- Multiple files: the drawer shows a row of tabs at the top, one per file. Click a tab to switch between files. The active file previews below.
The drawer stacks on top of any other side panel (like the Work items panel) you had open. Close it by clicking anywhere on the chat or by opening a different panel.
Downloading or sharing
The artifact preview in the drawer includes a download option so you can save the file locally and send it wherever it needs to go: as an email attachment, into a shared drive, back to the customer who asked for it. Downloads use the artifact’s real filename, so an Excel response keeps its .xlsx extension and name.
To share the work with a teammate, download the file and send it through your usual channel. Artifacts don’t have a direct “share link” from inside Circuit.
What artifacts can contain
The agent can produce artifacts in the formats it has write tools for, which typically include:
- Spreadsheets (
.xlsx, .csv): populated response files, crossover tables, inventory summaries, anything row-based.
- Word documents (
.docx): drafted proposals, RFP narrative sections, memos.
- PDFs: reports or formatted outputs the agent composes.
- Plain text and markdown: notes, summaries, generated instructions.
- Images (
.png, .jpg): snapshots or extracted visuals where supported.
The exact mix depends on how your agent is configured. If you need a format and the agent can’t produce it, ask: the agent will tell you what it can write, or suggest a close alternative.
Asking for the artifact you want
Agents produce better artifacts when you describe the output shape up front. A few patterns that work well:
- Name the format explicitly. “Write the result as an Excel file with columns: Competitor PN, Our PN, Description, Confidence” gives the agent a concrete target.
- Describe what the file is for. “This will be emailed to the customer” or “I’ll paste this into our proposal template” helps the agent calibrate tone and formatting.
- Request one artifact per output. If you need a response spreadsheet and a summary memo, ask for both explicitly. The agent can produce multiple files in a single reply, but only if you tell it to.
- Iterate. If the first pass isn’t right, say what to change: “Add a Notes column”, “Sort by confidence descending”, “Shorten the summary to one paragraph.” The agent can revise and post an updated artifact.
What artifacts don’t do
A few things to be aware of:
- Artifacts aren’t editable inside Circuit. The preview is read-only. To make changes, either ask the agent to revise it, or download the file and edit it in the appropriate application.
- Artifacts don’t sync back. Edits you make to a downloaded copy don’t flow back into Circuit. If you want the agent to work from your edits, re-attach the edited file to the chat and tell the agent what changed.
- Artifacts belong to the chat they were created in. They’re easy to refer back to inside that chat, but they aren’t automatically available in other chats. If you need the file again later, keep the download.
Example
You: Take the attached Globex parts list and produce a response spreadsheet with our matched part numbers, descriptions, and confidence. Keep the original row order.
Agent: [Posts a reply explaining the matches, with an artifact card titled “Globex crossover response” and subtitle “globex_crossover_2026-04-24.xlsx.”]
You: [Clicks the card. The artifact drawer opens showing the spreadsheet with all 48 rows filled in.]
You: Can you add a Notes column and flag the ones where the flow rate doesn’t exactly match?
Agent: [Posts a new artifact card with the updated file. Opening it shows the added Notes column with flags on the rows in question.]
Open the artifact; don’t only skim the chat message. The agent usually summarizes what’s in the file, but the file itself is the deliverable. Review the artifact before sending it to anyone else.
