v0.2.4 — March 2026

VLM Browser Navigation

• Vision-based clicking — see the screen, click anywhere (~2.5s)
• Native screenshot capture via CoreGraphics (~5ms)
• Full browser navigation — all links work in-place

Agent Browser

• vlm_screenshot + vlm_act WebSocket commands
• Pixel-accurate coordinate mapping for VLM grounding
• Multi-window support for agent tabs