Skip to content

PRODUCT.md

Latest commit

 

History

History
52 lines (30 loc) · 3.2 KB

PRODUCT.md

File metadata and controls

52 lines (30 loc) · 3.2 KB
Raw

Product

Register

product

Platform

web

Users

Knowledge workers lead: professionals who want AI working on their local files without uploading anything to the cloud, and who are not developers. They need plain language and interfaces that don't assume terminal fluency. Developers protecting proprietary code and researchers with sensitive data follow as secondary audiences; when priorities conflict, the knowledge worker wins.

Their context: sitting down at a desktop app to start or resume project work. Their job to be done: get an AI agent to do real work on their own files, safely and in one organized place.

Product Purpose

Cowork-Z is a local-first desktop workspace for autonomous AI agents. Each project lives in one workspace folder: files, sessions, and agent history together. Agents run sandboxed with folder-level permissions, on any of 12+ model providers, including scheduled automations that run unattended.

Success looks like Cowork-Z becoming the daily driver: the place where all project work starts, replacing scattered chats and folders. A user who opens it every morning, not one who tries it once.

Positioning

AI agents on your local files — private, organized, and ready to run. The three claims carry equal weight: your data never leaves your machine, every project lives in one workspace, and agents work autonomously within limits you set. Every screen should reinforce at least one of them.

Brand Personality

Friendly, approachable, simple. Warm and inviting: the tone lowers the barrier for non-technical knowledge workers. Capability is real but never performed; the app feels like a helpful colleague, not a machine room.

Reference points:

  • Claude Desktop / ChatGPT for the chat surface: clean message stream, readable markdown, calm surfaces.
  • Notion for the workspace feel: soft neutrals, friendly empty states, low-jargon UI.

Anti-references

  • Hacker terminal aesthetic. No dark-mode-only, monospace-everywhere, matrix-green developer cosplay; it alienates the primary audience.
  • SaaS marketing gloss. No gradient heroes, glassmorphism, or growth-hack patterns inside the app.
  • Enterprise admin gray. No lifeless gray dashboard sprawl: dense tables and chrome with no warmth.
  • Toy-like AI novelty. No sparkle emojis, mascots, or AI-magic theatrics that undermine trust.

Design Principles

  1. Approachable power. Knowledge workers first: plain words over jargon, sensible defaults over configuration. Advanced capability is available, never a prerequisite.
  2. The workspace disappears into the work. Daily-driver ambition means earned familiarity: consistent affordances, patterns users already know from Notion and Claude Desktop, no novelty friction.
  3. Show the agent's hands. Trust in autonomous agents comes from visibility: permissions, tool calls, progress, and findings are always legible, never hidden behind magic.
  4. Privacy you can see. The UI itself demonstrates local-first: folder scopes, permission prompts, on-device storage. Copy alone doesn't earn that trust.

Accessibility & Inclusion

WCAG 2.1 AA: 4.5:1 body-text contrast, visible focus states, full keyboard navigation, and reduced-motion alternatives for every animation.