Prox OS Internal Docs
PlatformArchitectureShell

Command Surface Competitive Analysis

Cmd+K should be measured against Raycast-class command products, not against a

Purpose

Cmd+K should be measured against Raycast-class command products, not against a basic website search box. This document defines what Prox OS should learn from Raycast, Alfred, Spotlight, Linear, Superhuman, and adjacent command surfaces while keeping the product direction larger than a launcher.

Cmd+K is the Global Command Surface for Prox OS. Its north star is not only "find and open"; it is summon, generate, open, connect, and execute inside a visible, auditable, data-sovereign OS Space.

Competitive Frame

Product or surfaceWhat users expectWhat Prox OS should learnWhat Prox OS should not copyProx OS edge
RaycastFast keyboard launcher, extension store, commands, snippets, AI helpers, and polished result previews.Command speed, compact density, extension distribution, local muscle memory, quick actions, and trustworthy command metadata.Treating commands as only local productivity utilities or hiding automation authority behind extension convenience.Space-aware summoning, app/agent visibility, permission previews, audit logs, and data-source wiring.
Alfred / LaunchBarLocal search, workflows, clipboard, snippets, and power-user automation.Zero-lag entry, fuzzy ranking, keyboard ergonomics, fallback search, and composable workflows.Local-only assumptions and opaque workflow permissions.Browser OS runtime context, app manifests, connector scopes, and future cloud/local hybrid execution.
Apple SpotlightUniversal system search with app, file, web, and setting results.Default system muscle memory, simple ranking, and low-friction entry.Becoming a shallow index without command verbs.Results can become windows, widgets, agents, Space layouts, and reviewable actions.
Linear command menuContextual product actions inside a focused workflow.Action verbs, scoped context, fast navigation, and predictable keyboard flow.Making commands only app-local.Cmd+K can combine Shell, current app, current Space, docs, agents, and registry results.
Superhuman command menuEmail-specific command speed and expert workflow shortcuts.Fast command phrasing, command confirmation, and low visual noise.Domain-specific narrowness.Prox OS can route commands across apps, data, connectors, and agents.
VS Code Command PaletteExtension-driven command provider model.Provider contracts, command discovery, keyboard-first work, and extension contribution points.Developer-tool-only mental model.App manifests and future agent registries can expose commands to non-code Spaces.
Slack shortcuts and workflowsTeam actions, workflow triggers, and app integrations.Clear verbs, app integrations, and workflow affordances.Chat-first command dependency.Cmd+K is a shell-level surface, not a message box.

Product Bar

BarRequirement
SpeedOpen instantly, keep focus in the input, support predictable keyboard navigation, and avoid expensive providers on first paint.
RankingPrefer exact app names, active Space context, recent windows, registered app manifests, and explicit command verbs before broad search.
PreviewShow what will happen before action: app, route, Space target, permission posture, disabled/future state, and execution risk.
Provider boundaryShell, app, Studio Engine, docs, search, agent, and connector providers should contribute through typed contracts rather than hardcoded route lists.
Permission postureConnect and Execute commands must surface scopes, cost, audit, and confirmation needs before mutation.
Space awarenessSummon and Generate should target the current Space or make the target Space explicit.
Failure handlingMissing manifests, unavailable connectors, disabled agents, and offline providers should degrade without breaking the Shell.

Verb-by-verb Differentiation

VerbRaycast-class baselineProx OS target
SummonOpen apps, windows, snippets, and extension views.Bring app windows, widgets, agents, layouts, or app groups into the current Space.
GenerateAI command creates text, snippets, or lightweight assets.Draft an app shell, agent workspace, Space layout, manifest, or reviewable branch proposal.
OpenLaunch apps, routes, files, and extension commands.Open registered apps, docs, routes, restored windows, Space resources, and platform tools through manifests.
ConnectIntegrations usually live inside extension settings or workflow builders.Create typed edges between apps, data sources, connectors, agents, and permissions with review.
ExecuteRun local commands, scripts, or extension actions.Execute only approved actions with permission preview, manual confirmation, audit logs, cost limits, and revocation.

Architecture Implications

AreaImplication
App RegistryOpen and Summon results must resolve through registered manifests rather than arbitrary route strings.
Agent RegistryFuture agent results need visible identity, scopes, budgets, run status, and audit posture.
Command providersProviders should be lazy, scoped, typed, and cancellable so Cmd+K stays fast.
SearchSearch can feed results into Cmd+K, but search should not own command execution.
PermissionsConnect and Execute need permission preview surfaces before they mutate data, spend budget, or call tools.
Audit LogSensitive commands should produce user-readable audit events.
SpacesCommand results need a target Space, active Space context, or a clear cross-Space action label.

v0.1 Guidance

DoDo not
Make future command capability visible through disabled or preview-labeled cards.Claim real agent execution, workflow automation, or connector mutation before it exists.
Keep existing app opening reliable and registry-driven.Hardcode a parallel app launcher list inside the command UI.
Use clear safety labels such as Requires permission, Coming later, and Audit log required.Hide risk behind exciting action verbs.
Treat Raycast as the speed and polish benchmark.Treat Raycast as the final product boundary.

Review Checklist

  • Does Cmd+K still open fast with no heavy provider blocking first paint?
  • Are app launches still manifest-driven?
  • Do disabled future actions explain why they are disabled?
  • Do Execute-like commands show permission, audit, and confirmation posture?
  • Does the command result make the target Space or active context visible?
  • Are provider errors contained without crashing the Shell?

On this page