Multimodal Input for HarnessAgent
Vercel has announced an update to its AI SDK library, officially introducing image input support for the HarnessAgent class across all existing adapters. This enhancement enables developers to pass images alongside text prompts when interacting with AI coding tools and automated task execution environments.
Standardized Message Schema
According to Vercel's documentation, users can submit images in parallel with text using the AI SDK's standard message format. The data structure utilizes a 'content' array containing both text parts ({ type: 'text' }) and file parts ({ type: 'file' }). Submitted image files are processed directly by the underlying vision model integrated into each respective platform.
Broad Adapter Ecosystem Support
The vision capability is enabled simultaneously across all 10 existing adapters in the HarnessAgent ecosystem:
- Claude Code - Cline - Codex - Cursor - Deep Agents - fx - GitHub Copilot - Grok Build - OpenCode - Pi
This unified implementation allows developers to build multimodal workflows without writing platform-specific handling code for each coding assistant or runtime environment.
Availability and Documentation
Vercel has published detailed documentation on configuring image inputs for HarnessAgent on the official AI SDK docs page. The company has not yet released specific details regarding file size limits, supported image formats, or any potential cost implications when making vision calls across different adapters.