Browse.
Capture.
Compound.

The landing screen. One field that searches your library or takes a URL, four quick-capture actions, live counts for every library type, your recent captures and whatever the agent is currently running. Built to get you browsing in one keystroke.

The actual browser: tabs, back/forward/reload, a URL field that falls back to web search, viewport presets and the Capture / Extract / Convert toolbar. The page itself runs in an isolated native view, so it never touches Nisaba's internals.

Turning on Extract makes the page hover-selectable — arrow keys walk the DOM to the parent, child or sibling of whatever you're pointing at.

The right-hand panel, in three tabs. Inspect shows the selection's name, screenshot, DOM summary, CSS custom properties, fonts, colors, detected dependencies and framework guesses — every detection carrying a confidence score and its evidence. Assets lists the images, icons, SVG and fonts found inside it. AI is where you choose an output profile and agent before converting.

Two captures, three ways: side by side, overlaid with an opacity slider, or as a pixel difference that paints every change in purple and tells you what percentage moved. Point it at a source reference and a later recapture to see what a site changed, or at a screenshot of what you built to see how close you got.

The difference is computed locally in a canvas, and pixel equality is never the goal — it is there to show you what moved, not to grade you.

The bottom strip. Every agent job appears here with live progress, streaming logs, cancel and retry. It survives a restart — a job that was running when the app closed gets reconciled rather than silently lost.

⌘K from anywhere. Jump to any library, reopen any tab, or type a URL to open it in a new one. It is the fastest path between "I want to look at that" and looking at it.

Every page review you have run. Each one keeps its pins — the note, the priority, the category, the element it points at and the file Nisaba matched it to — plus where the plan was exported.

Reopen the page to carry on, or export the plan again after the code has moved.

Links worth keeping, with tags, collections, notes and a status. Bulk add takes a pasted list — one URL per line, schemes optional, duplicates and blank lines dropped automatically. Broken links get flagged rather than deleted.

Every screenshot you've taken, grid or table, filtered by site, tag, date or viewport. Each one keeps its source URL, page title, scale factor and scroll position, with the annotation layer stored separately so the original is always recoverable.

The good stuff: reusable page regions typed as Hero, Pricing, Navigation, Form, Testimonial, Footer and so on. A section carries its screenshot, selector, sanitized HTML, matched styles, downloaded assets, accessibility metadata and a full trail back to the page it came from.

The Element Style Matrix. Run Detect elements on a page and Nisaba finds the buttons, inputs, checkboxes, cards, badges and tables, collapses visually identical instances so you get variants rather than repeats, and works out which pseudo-class rules the page actually declares.

Each state it finds — :hover, :focus, :active — is screenshotted for real by forcing that state through the DevTools protocol, so the matrix shows what the element does, not what you assume it does.

The tools rather than the designs: icon libraries, UI kits, font collections, repositories. Classified on save with package metadata, repository links and license where they can be found, and cross-linked to every site and section that uses them.

One record per domain — favicon, notes, tags, detected technology and every capture, section and bookmark you have taken from it. Recapture later and Nisaba will show you what changed.

A site's design language, extracted: color roles, type hierarchy, spacing and radius scales, breakpoints and observable motion. Exports as an editable design.md, tokens.json and site-profile.json. Observed values and inferred ones are always labeled differently — Nisaba never pretends a guess is a fact.

What the agent built. Each component keeps its files, framework, preview, the exact prompt profile used, the job that produced it and the sections it came from. Every generated version is preserved, so you can roll back.

Multi-section pages. Order saved sections into an outline, choose what to keep and what to genericize, and let the agent assemble the whole thing inside a folder you nominated.

The full history of agent runs, not just the ones currently going. Inputs, resolved prompt, adapter, streamed events, logs, output files, status and timings — kept so you can see exactly what produced a component months later.

A workspace is a root folder plus an output stack, prompt profile, agent and permission scope. Jobs can only write inside an approved workspace, and Nisaba shows you the command, the working directory and the permissions before the first write-enabled run.

Where your library, components and templates live on disk. Which agent CLIs are detected and authenticated. Sanitization defaults for saved HTML. Diagnostics — off unless you turn them on.