Beyond Text: Introducing Artifacts for Vectara Agents

Until now, most AI interactions were text-based. If you wanted an agent to analyze a report, you had to copy-paste the text or go through the process of indexing the document into a corpus just for a quick, one-off analysis. If you wanted it to look at a chart, you were out of luck.

With the release of artifacts, that changes today.

We are transforming Vectara Agents from text-only assistants into multimodal, file-aware systems. Agent sessions can now hold files, analyze images, convert documents (like PDFs and Word files) to markdown or structured formats, and create their own content (like reports or summaries) that persist throughout the conversation.

What Are Artifacts?

An artifact is a piece of content that is stored in the agent session. Artifacts are persistent throughout the session, which means:

1. Users can upload PDFs, spreadsheets, and images to a session.
2. Agents can read, analyze, and cross-reference these across multiple turns of conversation of the session
3. Agents can create new artifacts (like summaries, reports or code snippets) that are added to the same session, and the user can download later.

Note that the scope of artifacts are agent sessions, not the agent itself. This means that an artifact uploaded in a session is only visible for that session, not across other sessions of the same agent.

Artifacts are subject to size limits per file and total storage limits per session. While standard files are stored in the session, images are loaded directly into the conversation context via the image_read tool. For larger text-based files that exceed context limits, agents can read specific sections using line ranges with the artifact_read tool.

Artifacts have a configurable time-to-live (TTL). By default, artifacts are automatically deleted after a set number of days (configurable per account). Artifacts are also removed when the session is deleted. A user can control the retention period of a session by setting the session's time-to-idle parameter—when a session expires due to inactivity, its artifacts are deleted.

Required Tool Configurations

Critical Configuration Step:

Uploading a file creates the Artifact, but your agent cannot "read" the contents unless you give it the right tools. By default, an agent knows an artifact exists (it receives an ArtifactUploadEvent), but it acts blindly until you enable the specific tools below in the Vectara Console or API.

Real-World Use Cases

Financial Analysis Across Multiple Documents

• Upload Q3_Financials.pdf and Q4_Projection.xlsx to the agent session
• The agent receives artifact upload events for both files
• With document_conversion added to the agent, it converts the PDF to extract financial data
• The agent uses artifact_read to access spreadsheet content
• It cross-references Q3 actuals against Q4 projections, identifies variances, and provides analysis

Visual Debugging with Screenshots

• Upload error_screenshot.png showing a checkout page error
• Ask "What does this error mean and how do I fix it?"
• The agent uses image_read to load the image into conversation context
• It visually analyzes UI elements, reads error messages, identifies the error type
• The agent suggests fixes based on what it sees in the image

Document Conversion and Summarization

• Upload Technical_Manual.pdf
• Ask the agent to "Convert this to clean Markdown and generate a simplified 'Quick Start' guide"
• The agent uses document_conversion to transform the PDF into markdown, creating a new artifact
• It uses artifact_read to access the converted content
• The agent uses artifact_create to generate a Quick Start guide artifact
• Both the converted manual and summary remain in the session workspace for download

Developer Guide: API Implementation

Here is how to implement Artifacts in your application.

1. Upload a File to an Agent Session

Upload files alongside a message using multipart form data.

Or in this other example, you can see how once the artifact has been uploaded, it becomes part of the session context and can be used by subsequent events without re-uploading it.

Step 1: Upload file

Step 2: Later in the same session, reference that artifact in a new event

2. List Artifacts in an Agent Session

See what is currently stored in the session's workspace.

3. Retrieve Content of an Artifact

Download an artifact (including agent-generated artifacts).

Ready to Build?

Artifacts provide the foundation for sophisticated, file-heavy AI workflows that were previously impossible. Whether you are analyzing contracts, interpreting charts, or automating report generation, your agent is now ready to help.

Get started today by configuring your agent tools in the Console and uploading your first document.

For full details, visit the Vectara API Documentation.