Features
Everything Agentation can do
Annotation modes
Click the tabs below to see examples of each annotation mode:
"Simple can be harder than complex: You have to work hard to get your thinking clean to make it simpl. But it's worth it in the end because once you get there, you can move mountains."
Select text to annotate typos, content issues, or copy changes. The quoted text is included in the output.
Toolbar controls
- Pause • Pause page motion to annotate specific states
- Visibility • Toggle annotation markers on/off while working
- Copy • Get structured markdown for AI agents
- Clear • Remove all annotations
- Send Annotations* • Send annotations to your configured webhook
- Layout Mode • Place components and rearrange page sections
- Settings • Configure output detail, marker color, and behavior
Drag the toolbar to reposition it. Hover a marker to preview the feedback, then click to edit. You can change marker clicks to delete in settings; right-click always opens the editor.
*Only visible when webhooks are enabled and auto-send is off.
Marker types
Different annotation modes use different marker styles.
- 1Single element or text selection
- 1Multi-select or area (always green)
Smart identification
Agentation automatically identifies elements in a way that’s useful for code search. This makes it easy for agents to grep for the exact element in your codebase.
Benji Taylor
@benjitaylorhead of design @x & @spacexai
Buttons and links are named by their text content.
Computed styles
View the computed CSS styles for any element directly in the annotation popup. Expand the collapsible section to see relevant properties like colors, fonts, and spacing.
Click the chevron to expand computed CSS styles for the selected element. Useful for debugging styling issues or communicating design specs.
React component detection
Agentation detects React component hierarchies automatically. When you hover over an element, you’ll see the full component tree, making it easy for AI agents to locate the exact component in your codebase.
Hover over any element to see its React component hierarchy.
Toggle React detection on/off in settings. The detection mode adapts automatically based on your output format:
- Compact: No React data, so output stays minimal
- Standard: User components only, with framework internals hidden
- Detailed: Only components that correlate with CSS class names
- Forensic: Every component, including framework internals
Layout mode
Press L to enter layout mode. A component palette appears with 65+ types you can drag onto the page. You can also grab existing sections and reposition them. Changes sync to connected agents via MCP in real time.
- Place components: Drag from the palette to add new elements
- Rearrange sections: Grab and drag existing elements into position
- Wireframe new page: Toggle wireframe mode to fade out the current page and sketch a layout from scratch, with an opacity slider and a purpose field for context
- Agent sync: Changes appear as annotations with
kind: "placement"orkind: "rearrange"
Placements are spatial hints, not pixel-perfect specs. The agent receives component type, approximate position, and size relative to the viewport, then translates that into real CSS. Results will vary depending on the agent and your codebase.
Keyboard shortcuts
| Key | Action |
|---|---|
Cmd+Shift+F / Ctrl+Shift+F | Toggle feedback mode |
Esc | Close toolbar or cancel |
L | Toggle layout mode |
P | Pause/resume animations |
H | Hide/show markers |
C | Copy feedback |
S | Send annotations |
X | Clear all annotations |
Single-key shortcuts work while the toolbar is open, and all shortcuts are disabled when typing in an input field. Cmd+Shift+F and Esc also work while it is closed.
Agent sync
With MCP integration and the Annotation Format Schema, annotations become a two-way conversation. Agents can query, respond to, and manage your feedback:
Agents can acknowledge, ask questions, resolve with summaries, or dismiss with reasons.
Settings
The icon lets you customize Agentation to fit your workflow.
Choose how much detail to include in your output.
Fit your app
Pick targets inside open shadow roots and same-origin embedded pages. For a modal or popover, pass its content element to portalContainer so feedback stays within the host’s focus boundary.
Use an app name and stable data attributes to make feedback easier to identify. Copy full feedback or just source paths, classes or an attribute value. Add your own editor action when a source location is available.
Limitations
- Desktop only: requires a desktop browser
- Per-page storage: localStorage persists 7 days. Hash-based routes can opt into separate storage. Use the MCP server to share feedback with your agent.
- Layout changes: same-origin single-element frame markers track child scrolling. Arbitrary page reflows and saved group geometry are not continuously tracked.
- No screenshots: output is text-only
- Animation pause: freezes most animations and videos on the page. Some third-party animation libraries may not fully pause.
- React 18+ only