Skip to content
Type
design system
Verdict
Useful
Licence
Conditional licence, proprietary paidproprietary-paidProprietary / commercial; free tier for personal use only, paid tiers grant commercial use, no OSS licence or redistribution
Agent channels
llms-txt, api
Pricing
freemium
Reviewed

← Atlas · Topics: components, agents-and-prompts, ai-interfaces

Kobra ​

Type
design system
Verdict
Useful
Licence
Conditional licence, proprietary paidproprietary-paidProprietary / commercial; free tier for personal use only, paid tiers grant commercial use, no OSS licence or redistribution
Agent channels
llms-txt, api
Pricing
freemium
Reviewed

What it is ​

Kobra is a component system purpose-built for AI and agent product interfaces. It includes specialized pieces like a message scroller that sticks to the bottom of chat containers with a "N new messages" jump button, plan cards for approving or rejecting agent actions (with ⌘Enter shortcuts), inline citations with hover-revealed sources, file diffs with per-hunk accept/reject controls, input OTP fields, multi-select, select components that adapt to touch vs. pointer input, item rows, toggle groups, and chat-specific media pieces.

When to open it ​

Open Kobra when designing interfaces for AI assistants, agent dashboards, or chat applications. It's especially useful if your product needs specialized controls for reviewing agent outputs, comparing diffs, or managing structured input.

Most useful ​

  • Agent-aware components like plan cards and citation links that anticipate workflows around LLM outputs
  • llms.txt index and markdown API (/r/<name>.md) with exports and package info—agents can read these directly
  • Keyboard shortcuts baked into components (e.g., ⌘Enter on plan cards)
  • Responsive controls that detect pointer type and adapt (native select on touch, custom on desktop)
  • Live examples at /components/<name> with interactive previews

Using it with agents ​

Kobra publishes detailed markdown documentation per component, accessible at predictable URLs. You can point Claude to these pages or paste component names into agent prompts; the system is designed to be agent-friendly. Keyboard shortcuts and approval workflows map naturally to LLM outputs.

Watch out for ​

  • Most restrictive licensing of the sites in this atlas—ideas only, never copy code or component names
  • Free tier is personal use only; commercial projects require a paid license
  • No open-source license or redistribution rights; you cannot fork or reuse code outside the commercial product
  • Dependencies include input-otp, motion, Embla, Vaul, kmenu, and Tabler/Hugeicons—check compatibility with your stack

Reusable ideas ​

  • Chat containers that auto-scroll to new messages but allow manual scroll-up
  • Action approval flows with keyboard shortcuts for efficiency
  • Inline citations linked to their sources (visible on hover)
  • File diff UI with fine-grained accept/reject controls
  • Message type indicators (user, assistant, error) as visual badges
  • Plan cards for human-in-the-loop agent validation

shadcn/ui, Aceternity UI, The Component Gallery