Changelog

All notable changes to AInotate.

0.1.0 - 2026-10-07

First public release.

Added

• Annotated screenshots from a JSON spec: numbered steps, boxes, labels, Skitch-style tapered arrows with a soft shadow, highlight, spotlight, text notes and solid redaction.
• ainotate package (src layout, AGPL-3.0-or-later) with the ainotate command and python -m ainotate. Rendering is pure: it returns the image and warnings.
• Content-aware placement: a map of the page content (text, icons, lines); labels go to free space near their target and are planned together; arrows are priced by length and by what they cross, curve around text and bend gently when long; step badges take the corner that covers the least. A warning when a label has to cover text. Four arrow styles: skitch, curved, straight, line.
• grid, zoom and info to read exact positions at full resolution; one units rule with scale.
• ainotate shoot: URL plus marks with target in one call: capture, run actions, measure, redact, annotate, save. Presets laptop, wide and phone.
• Web capture for any browser: own Playwright (Chromium, Firefox, WebKit) or attach over CDP to a running Chromium (Chrome, Edge, Brave, Arc, BrowserOS neo) with per-tab emulation that is always restored. Targets by text, role or CSS selector; actions click, fill, wait, scroll_to, wait_ms; overlay hiding; same-origin iframe offsets; retry over HTTP/1.1 when a navigation stalls silently.
• OCR text targeting on any image: ainotate locate and "target": {"text": ...} on marks, with Apple Vision, Windows.Media.Ocr or Tesseract. Exit 5 for ambiguous text, 6 for text not found (closest matches listed).
• Automatic privacy redaction: emails, phones, cards (Luhn), IBANs (mod 97), API keys, JWTs, tokens, password and personal-data fields, custom regexes and allow lists. On by default for web shots (live DOM, input values included), opt-in on images (OCR). Solid blocks only.
• Marks magnify (loupe), click (ripple and pointer), keys (keycaps), decorative blur and pixelate (refused for data marked sensitive); overlap and crossing warnings.
• Frames: auto or preset gradient backgrounds, rounded corners, soft shadow, browser or window chrome with the page address, aspect presets for social posts.
• Export: step guides in Markdown, HTML and PDF; APNG and GIF animations; before/after plates; copy to the clipboard.
• Desktop capture: screen, region, window list and window capture, clipboard image, on macOS, Windows and Linux. Windows: explicit ctypes signatures, DWM frame bounds, per-monitor DPI, multi-monitor origin.
• ainotate doctor: parallel checks of Python, Pillow, font, output folders, Playwright, CDP, screen recording permission, window listing, OCR, clipboard and MCP, each with the exact fix for the current OS; --json.
• MCP stdio server (ainotate mcp) for Claude Desktop and other clients: tools for shoot, capture, annotate, preview, locate, grid, zoom, desktop capture, guides, compare, animate, clipboard and doctor; JSON results with small previews; spec resource and two prompts.
• CLI flags --draft, --debug, --copy, --json, --privacy; exit codes 0 ok, 1 save, 2 spec or usage, 3 render, 4 capture or missing backend, 5 ambiguous text, 6 not found.
• Configurable output: env, ~/.config/ainotate/config.toml, defaults; never overwrites.
• Agent skill in skill/ that drives the CLI, with references for the spec, browsers and the MCP server.
• Bundled Inter Bold (SIL OFL 1.1), so labels render the same everywhere, diacritics included.

Platform status

• macOS verified. Windows and Linux are experimental: implemented and unit-tested with mocks, not yet verified on real machines.