Add Image File Support as a Knowledge Source

[PeanutSplash]

Description

Feature Request: Image File Support in Knowledge Sources

Problem

Currently, the knowledge agent only supports text-based sources (GitHub repos,
YouTube transcripts, custom APIs). Many real-world knowledge bases contain
valuable information embedded in images — diagrams, charts, screenshots,
scanned documents, and architecture diagrams — which the agent cannot process.

Proposed Solution

Add support for image files (.png, .jpg, .jpeg, .webp, .gif) as a
knowledge source type, enabling the agent to:

1. Ingest images from configured sources (GitHub repo folders, URLs, uploads)
2. Extract text/content via OCR or a multimodal LLM
3. Store extracted content in the sandbox snapshot so existing grep/cat
   tools can search it like any other text file
4. Preserve image metadata (filename, alt text, captions) alongside content

Use Cases

• Searching architecture diagrams or flowcharts stored as images
• Indexing scanned PDFs or documentation screenshots
• Querying infographics and charts containing embedded text
• Supporting docs where visuals are the primary communication medium

Additional context

No response