Sessions

Sessions are conversational threads where you interact with the VisoWork AI agent to analyze visual content.

Starting a Session

Start a session by typing a message or uploading a file from the project dashboard. The AI agent automatically selects the best vision tool for your request.

Example prompts:

> Describe this image: /path/to/photo.jpg
> Detect all safety hazards in this construction video
> Extract text from this document scan
> Compare the color grading between these two clips

Session Interface

Message Types

Sessions support several message types:

• User messages — Your text and file uploads
• Agent responses — Analysis results with structured data
• Tool calls — Shows which API method the agent used (e.g., vision.describe, vision.detect)
• Tool results — Execution status, duration, and result summary
• Thinking indicators — Shows when the agent is processing

Rich Content

The agent can render rich content in responses:

• Annotated frames — Images with bounding boxes and labels overlaid
• Hazard cards — Timestamped findings with location, image crop, and description
• File attachments — Inline display of uploaded images and videos with metadata
• Code blocks — Structured JSON results

Session Controls

Session Status

• Running — Analysis is in progress (pulsing indicator)
• Completed — All analysis steps are finished

Sandbox Mode

For complex analysis tasks, the agent may enter Sandbox mode — a full compute environment with Python, FFmpeg, and OpenCV. The sandbox indicator appears in the session header with a "Stop Sandbox" button.