Artifacts
Let your agent produce rich, structured output - documents, code, images, spreadsheets - that end users can read, edit, and download.
Artifacts
An artifact is a piece of rich content your agent produces alongside (or instead of) a plain chat reply. Where a chat answer lives in the message stream and scrolls away, an artifact opens in its own panel so end users can read it carefully, edit it, copy it, or download it.
Use artifacts when your Flow's job is to make something - a report, a draft, an image, a spreadsheet, a snippet of code - not just to chat.
Artifact types
FormWise supports several artifact kinds. Your agent picks the right one based on what it's producing.
| Type | What it is | Example |
|---|---|---|
| Text | A formatted document with headings, lists, and inline edits | A drafted email, a research brief, a proposal |
| Code | A code block with syntax highlighting | A SQL query, a snippet of HTML, a Python script |
| Image | A generated image | A logo concept, a hero illustration |
| Sheet | A tabular grid you can read like a spreadsheet | A cleaned-up CSV, a pricing comparison |
| Flow preview | A live preview of another FormWise Flow the agent built or referenced | A form mid-build |
| Build progress | A streaming status display for long-running Flow builds | Step-by-step "what the agent is doing" updates |
Note - Type availability evolves. The agent decides which type fits the request; you don't have to pick in advance.
What end users see
When your agent emits an artifact:
- The artifact streams in inline in the chat so the user sees it being generated.
- Once complete, it opens in a side panel beside the chat. The chat stream continues to show a compact reference.
- The user can scroll the conversation freely while the artifact stays pinned.
- For text artifacts, the panel is a rich editor - users can select a passage and ask the agent to rewrite just that part, leaving the rest intact.
[Screenshot: chat on the left, expanded text artifact on the right]
Enabling artifacts
Artifacts are enabled per Agent node.
- Open your Flow in the workflow builder.
- Click the Agent node where you want artifacts.
- In the agent settings panel, find Artifacts under the Tools section and turn it on. (This is on by default for new Agent nodes.)
- Save and test.
For Agents, artifacts are on by default - no toggle needed.
When to use artifacts vs. plain text
| Situation | Use... |
|---|---|
| The user asked a question | Plain text reply |
| The user asked you to draft, write, generate, or build something they'll want to copy or keep | Artifact |
| The output is several paragraphs of structured content | Artifact (text) |
| The output is a one-liner or a list | Plain text |
| The output is code the user will paste somewhere | Artifact (code) |
| The output is data the user will paste into a sheet | Artifact (sheet) |
| The output is a follow-up clarifying question | Plain text |
A good heuristic: if the user will want to keep it, copy it, or edit it, it's an artifact. If it's part of the conversation, it's a reply.
Tips
- Trust the agent's judgment. Modern models pick artifacts vs. text well. If you find your agent producing artifacts when it shouldn't (or vice versa), tweak the agent's system instructions rather than disabling the feature.
- Title things clearly. Encourage the agent (via instructions) to give each artifact a short, descriptive title. End users will see these in the panel header.
- Combine with chat attachments. A common flow: the user uploads a doc, the agent reads it, then produces a polished artifact based on it.
- Voice and artifacts coexist. When voice mode is on, the agent still emits artifacts to the screen - voice handles the conversation, the panel handles the deliverable.
Limitations
- Image generation depends on a connected image-capable model. If you haven't configured one, image artifacts won't appear - the agent will fall back to a description.
- Very long text artifacts may take a few seconds to fully stream in. The panel shows progress while it builds.
- Artifacts produced by the agent live with the conversation. Downloading the artifact gives the end user a copy; deleting the conversation removes it.