Peek#

Peek is a web user agent application designed for using the web where, when and how you want.

Today's browsers are one-size-fits-all applications, cramming the vast universe of user needs across an unimaginably large web into in an unmodifiable tabbed-window design.

We often use the web with a specific goal in mind - that goal should drive the design of the interface of the web user agent. We should be the driver, creator, maker, shaper of that design and interface.

Peek is designed to allow you to experiment and create flows for task alignment in interfaces that are personal. It aims to make it easy to create new user interface shapes for the web which fit your need in the moment.

Peek is a workbench for the web. Peek is a window manager for the web. Peek is what you want it to be.

WARNING!#

PEEK IS NOT A BROWSER

Peek is not a browser, and will never be a browser in the way you are probably familiar with. There are no tabs, and no windows in the tabbed-browser-like sense of them. Peek likely does not have many other of many details we are used to in web browsers, but do not notice until they are missing. Peek may be most useful to you if you view it as an entirely different type of application than a traditional web browser.

PEEK IS A CONCEPT PREVIEW

Peek is not safe for daily use yet! It is a proof of concept. Do not use it for anything critical. Peek does not have the same security approach as traditional web browsers, and its security model and security user interface have not yet been determined. Peek has not had a security audit.

Core#

Built-in capabilities that form the foundation of Peek.

• IZUI Navigation — Inverted Zooming User Interface. ESC-key-based navigation model that provides consistent "zoom out" behavior across all interfaces. Enter from anywhere, ESC to go back — predictable at every level.
• Window Management — Heterogeneous windowing: chromeless modals, overlays, slides, workspaces and content windows, each with role-specific behavior.
• Feature System — Features are not extensions of a monolithic core — they are the functionality, built on a minimal foundation. Each feature declares its own commands, shortcuts, and settings. Features can be eager, lazy-loaded, or fully declarative.
• PubSub Messaging — Cross-window event system for communication between features and windows.
• Data Persistence — SQLite datastore with schema management, history tracking, and cross-device sync support.
• peek:// Protocol — Custom protocol for internal navigation and cross-origin access.
• Settings — Per-feature configuration with live reload. Cmd/Ctrl+, or tray icon.

Features#

Peek uses the language of "features" over "extensions". "Extend" implies a larger core that you cannot change. Peek inverts that to small core, with as much changeable as possible. Peek features are modular - they are not extending core functionality, they are the functionality, built on a small-as-possible core.

• Commands — Global-hotkey-activated graphical command bar (GCLI) for opening pages and features, executing commands, chaining commands. Opt/Alt+space (global) and Cmd/Ctrl+k (local)
• Groups — Categorize, recall and interact with groups of pages. Groups are tags that've been "promoted" intentionally to a group. Groups are also an app-global "mode" (which isn't entirely implemented yet...) Cmd/Ctrl+G
• Peeks — Global-hotkey-activated modal chromeless web pages for quickly glancing at or interacting with pages. Opt+0-9
• Slides — Global-hotkey-activated modal chromeless web pages which slide in from any screen edge. Opt+Arrow Keys
• Editor — Markdown editor with outline sidebar, live preview, vim mode and an excellent folding approach. Cmd/Ctrl+E
• Search — Search across all items by text and tags.
• Tags — Tag, untag, and browse items by tag. Tag visualization and management.
• Tag Actions — Define custom actions triggered by tags, enabling automated workflows.
• Web Search — Search the web with engine discovery — Google, DuckDuckGo, Kagi, Bing, Wikipedia, and more.
• Windows — Full-screen window switcher overlay showing all open windows as cards. Cmd/Ctrl+Shift+W
• Pagestream — Chat-like vertical navigation interface for browsing history and opening and viewing pages. Cmd/Ctrl+Shift+P
• Entities — Automatic extraction of people, places, organizations, and events from visited pages.
• Scripts — Periodically execute scripts against web pages in the background to extract data, with change notifications.
• Files — Open local files, CSV conversion, markdown formatting, save operations.
• Feeds — RSS/Atom feed reader.
• Lexicon Studio — Lexicon Studio: AT Protocol identity and data browser/editor.
• Widget Sheets — Widget sheets with freeform card layouts hosting webviews.
• Timers — Countdown timers, alarms, stopwatches, and interval timers.
• Sync — Cross-device synchronization of items, tags, and groups.
• HUD — Always-on-top heads-up display showing current state and mode. Ctrl+H

Coming:

• "native" web apps — using Peek as a way to "install" web pages on the local device as separate applications
• "Peeklets" HUD — select parts of pages to add to a collection rendered as an overlay

Keyboard Shortcuts#

Global (work from any application)#

Application#

Web Page Navigation#

Design#

Some thoughts driving the design of Peek

Core

• Feels like home: trust, comfort, control
• Continuous instances of magical mind-reading
• Sleep at night because no idea or anything you saw is ever lost
• Create, save, classify at the speed of thought

What makes a home / kitchen / workshop

• Everything is right where you need it, b/c you control what is where
• When you know what is where, you can make things with less frustration
• You generally know what’s happening - who’s around/coming/going, etc, not too much surprise

What makes magical mind-readingness

• Frecency everywhere all of the time
• User actions are training data for sorting
• The thing you were thinking of is already at the top of the list

What makes awareness

• Actions tracked at the core
• Metrics generated, rollups displayed, synthesis emergent
• Task context with you at the center (vs asking a companion robot - disempowering!)

Synthesis

• Frecency + adaptive matching gives experience/feeling of magical mind-readingness
• Ability to customize/create/generate interfaces gives the comfort of home

The rules

• Files > arcane/opaque boxes/formats
• Metadata can be weird non-file, as long as consistent
• External systems require consent to touch my stuff

Key capabilities from action/application history chain + addressibility

• Record/replay
• State feedback loops
• Observability

Interface spectrum

• Consistent predicable interfaces are easy, off-the-shelf
• Weird/generative possible via power trio: backgrounding, wm apis, web surface

Difference from browsers

• Web user agents should be bounded by the user, not browser vendor business models
• Windows and tabs should have died a long time ago, a mixed metaphor constraining the ability of the web to grow/thrive/change and meet user needs
• Security user interface must be a clear articulation of risks and trade-offs, and users should own the decisions

Task alignment#

Many user tasks on the web are either transient, chained or persistent, data oriented, or some mix of those. Neither the document-oriented nor application-centric web meets those needs. Traditional browser makers can't meet those needs well, for many reasons.

Characteristics of how we use the web, that are not addressed in contemporary web browsers:

• transient
• chained
• persistent
• data-centric
• archival / evidential

Escape IZUI#

TODO: articulate the escape-to-leave aspect, eg you can peek from other applications and ESC to go back to exactly where you were without breaking the task flow.

Escape is an inverted zooming user interface (IZUI) design for a flexible window manager that makes possible a web user agent application than can have multiple entry points and a heterogeneous windowing ecosystem.

IZUI vs ZUI

• ZUIs navigate by starting from a known root and user navigates by zooming ever further in, and then back out
• Escape can enter a window stack at any point, and via a variety of methods, often from outside the application
• Instead of navigating by zooming in, all interfaces can zoom out to go back, using the Escape key
• This design allows unbounded and diverse entry points, but with predictable behavior
• Regardless of the entry point, the user always has a consistent path to familiar ground

Escape navigation model

• navigation base can start at any level in stack
• forward navigations are added on top of stack
• backwards navigations walk the stack in reverse up the tree to the root

Architecture / Implementation#

About this space:

• Web pages can themselves be navigators of the web
• Embrace the app-ness of the web platform, as a way to efficiently access the document-ness
• Decouple html+js+css from http+dns+ssl - not entirely, but that trust+security model is not a required starting point
• Javascript is ok here

Peek is designed to be modular and configurable around the idea that parts of it can run in different environments.

For example:

• Planning on a mobile app which syncs and runs your peeks/slides/scripts
• I'd like to have a decentralized compute option for running your scripts outside of your clients and syncing the data
• Want cloud storage for all config and data, esp infinite history, so can do fun things with it

Feature architecture#

Peek inverts the traditional browser model. Instead of a large monolithic core with an extension API bolted on, Peek has a minimal core with features that are the functionality. Features are not extending something — they are the thing.

A modular feature system for achieving "personal web workbench" requires a few things:

• UI modularity requires OS-level window capabilities beyond what the web allows today (also a baby step towards a minimal OS user interface)
• Data harvest/transform/process/publish requires a method of moving data between features (web apps) locally, cf Web Actions/Intents/Applets, MCP, pubsub, MQTT etc
• Portable ways of accessing network, storage and compute

The current implementation has gone through a few iterations:

• first proof of concept was all Electron — privileged JS
• second experiment moved each feature to a separate web app running in own window scope, with access to smallest possible custom API, with one main web app loading and orchestrating the others, using pubsub for cross-app communication
• third and current implementation bundles all features into one web app, with access to smallest possible custom API for platform-level capabilities

Features are loaded into the custom peek scheme, which provides access to a few special APIs noted in the next section, allows cross-origin network access and other things.

The API surface area is small and portable — features built on it could run in other environments with minimal adaptation.

Feature API#

Features are web apps running in the peek:// scheme. The question driving the API design: what's the minimum capability set a web app needs to build these features?

The API is exposed as window.app in the feature's execution context:

• Windows — open, close, modify, and track windows with role-based behavior
• Shortcuts — register global and local keyboard shortcuts
• PubSub — publish and subscribe to messages across features and windows, with scoping (self, global, window)
• Commands — register commands for the command bar, with parameter modes and chaining
• Datastore — SQLite-backed persistent storage with schema management
• Escape — register ESC key handlers for IZUI navigation
• IZUI — query and interact with the navigation state machine
• Files — open/save dialogs, read/write by path
• Settings — per-feature preferences with schema validation
• Themes — access current theme, dark mode state
• Sync — cross-device synchronization
• Profiles — user identity and profile management
• Net — network requests (fetch with cross-origin access)
• Session — session state and restore

The API surface is intentionally small and portable — features built on it could run in other environments with minimal adaptation. See docs/api.md for the full reference.

Desktop App#

Proof of concept is Electron. By far the best option today for cross-platform desktop apps which need a web rendering engine. There's really nothing else remotely suited (yet).

User interface:

• the built-in features are all modal chromeless web pages at this point
• settings UI uses custom sidebar navigation with dark mode support

TODO

• Need to look at whether could library-ize some of what Agregore implemented for non-HTTP protocol support.
• Min browser might be interesting as a forkable base to work from and contribute to, if they're open to it. At least, should look more at the architecture.

Contribution#

• In proto stage
• All dragons, no promises

Mobile#

There's a LOT already done here and I haven't even begun to describe it yet.

Server/Cloud#

Server component is a very passive sync+store service right now.

Runs on Railway, I use it daily with few problems (that I know of).

History#

In working on Firefox and related things at Mozilla from 2006 - 2019, there were a few specific initiatives which best aligned with my needs as a user on the web:

• The Awesomebar: infinite history + personalized local search index
• Ubiquity: Natural language commands + chaining
• Jetpack: The Mozilla Labs version - web-platfrom-centric extensibility
• Panorama: née TabCandy, web pages as groups instead of tabs in windows

A few others which were in the right direction but didn't achieve their optimal form:

• Greasemonkey
• Microsummaries
• Contacts extension

The first version of the Peek application has some bits of each of these, and the original Peek browser extension.

Peek browser extension#

Peek was a browser extension that let you quickly peek at your favorite web pages without breaking your flow - loading pages mapped to keyboard shortcuts into a modal window with no controls, closable via the Escape key.

However, as browser extension APIs became increasingly limited, it was not possible to create a decent user experience and I abandoned it. You can access the extension in this repo in the extension directory.

The only way to create the ideal user experience for a web user agent that Does What I Want is to make it a browser-ish application, and that's what Peek is now.

Automated Summary (somewhat outdated, but some correct stuff too)#

Developer Quick Start#

# Requirements: Node.js 24+
nvm use 24

# Install and run
yarn install
yarn debug        # Development mode with devtools
yarn start        # Normal mode

Look at package.json for a long list of commands for each of the back-ends and testing.

See DEVELOPMENT.md for full development guide.

Architecture#

Peek supports multiple backends sharing the same storage/api approach:

peek/
├── app/                    # Renderer (backend-agnostic)
├── features/               # Built-in features
├── backend/
│   ├── electron/           # Desktop (primary)
│   ├── tauri/              # Desktop (Rust alternative)
│   ├── tauri-mobile/       # iOS/Android
│   └── server/             # Sync server (Node.js/Hono)
└── docs/                   # Documentation

Documentation#

Tests#

~2,000 tests across 69 files, using Node.js test runner, Playwright, and Cargo test.

Electron Unit (TS) — Backend logic: entities extraction, session management, IZUI state, cmd palette logic, cmd chaining, OAuth bridge, display watcher, noun registry, shortcuts, datastore, completers, tag affordances, page loading.

Unit (JS) — Frontend logic: cmd state machine, groups, name validation, tags search, editor outline/preview, page widgets, item queries, Chrome API polyfills (storage, alarms, events, navigation, IPC, offscreen).

Playwright Desktop — End-to-end: smoke tests (window lifecycle, navigation, IZUI escape protocol, cards, groups), history, cmd state machine, web search, HUD, local search, pagestream, page widgets, groups context, page navbar.

Playwright Components — Web component rendering and behavior tests for shared UI components.

Playwright Editor — Editor pane layout and code folding tests.

Playwright Extension — Browser extension options page and popup UI tests.

Extension Backend — Browser extension logic: datastore, tabs, history, bookmarks, sync, profiles, environment.

Server — Sync server: API endpoints, backup, migration.

Schema Fidelity — Validates schema consistency across desktop, mobile, server, and browser extension datastores.

Sync Integration — Cross-device sync: integration tests, E2E tests, three-way merge, version compatibility.

Lex — Lexicon feature: UI logic and background processing.

Background test commands: yarn test:electron:bg runs Playwright headless in background (logs to /tmp/test-electron.log).

Notes#

Design thinking, research, and implementation notes in notes/:

License#

MIT