Vision — Iris

Iris is the image-evidence tool for agents. Geometry is the default. Text is opt-in.

• Identity: package @sylphx/iris, bin iris, MCP io.github.SylphxAI/iris, site https://sylphxai.github.io/iris/.
• User: an agent that must read a screenshot, UI capture, form, diagram, chart, or photo and cite what it measured.
• Job: return dimensions, format, and metadata by default, and local Tesseract OCR only when requested.
• Promise: dimensions, hashes, and pixel bounds come from the file. OCR is labeled, scored, and omitted unless you ask. A missing Tesseract binary is a gap.
• Defaults: fast, or no profile, returns geometry and metadata. quality or include_ocr runs local Tesseract. Explicit include_ocr: false wins. GPS is always redacted.
• Boundaries: Iris owns local image facts. It does not own image generation, generic vision chat, photo galleries, or video timelines.