Skip to content

Latest commit

 

History

History
60 lines (40 loc) · 3.56 KB

VISION.md

File metadata and controls

60 lines (40 loc) · 3.56 KB

Vision

"Civilization advances by extending the number of important operations which we can perform without thinking about them." — Alfred North Whitehead

What screenpipe is

Context infrastructure for AI agents. The ambient layer between humans and their digital work.

Screen is the universal interface — 10M bits/second, the highest-fidelity signal of human intent. During work, the screen contains all the millions of software that ate the world and are being eaten by AI right now. Screenpipe captures that context locally and makes it available to AI.

Not a memory tool. Not an assistant you prompt. An ambient automation layer that works in the background with zero prompting.

Why we exist

Every AI interaction today requires stopping work, translating intent into a prompt, and waiting. That's a tax on every interaction. Your screen already shows exactly what you're doing — the context is right there.

We build the layer that gives AI full context of human work so it can act autonomously. Recording + AI = ability to clone human digital work at high fidelity.

Where this goes

  1. Now: Memory. Make desktop memory work so well people can't live without it. Record, Rewind, Ask — three verbs, nothing else.
  2. Next: Context layer for AI agents. Open API so any AI agent can query your screen history. Every AI agent needs to know what the user is doing, act on that context, and trigger without prompts.
  3. Later: Every sensor. Screen is sensor #1. Then cameras, rooms, spatial memory. Local-first makes it viable — "no internet required" level clear.

Product principles

  • Stability over features. Users who stay are obsessed. Users who leave hit bugs. Fix what's broken before building what's new.
  • No feature creep. Every feature must serve Record, Rewind, or Ask. If it doesn't, it doesn't ship.
  • Respect the user's machine. CPU, memory, disk — screenpipe runs 24/7 in the background. Performance is not optional. Target: <20% CPU, <3GB RAM on release builds.

Engineering principles

  • Ship daily. Small, focused changes. Every commit should be deployable.
  • Fix the funnel before adding features. Permission loss, onboarding drops, version fragmentation — these kill growth silently.
  • Local-first always. Data never leaves the device unless the user explicitly opts in (cloud sync, cloud archive). Encryption is zero-knowledge.
  • Cross-platform. If it doesn't work on macOS, Windows, and Linux, it's not done.
  • Open source by default. Trust is earned through transparency.

Design voice

  • State facts. No marketing fluff.
  • No emoji in the product. No exclamation marks. Remove what's unecessary.
  • Black and white. 1px borders. Sharp corners. No shadows, no gradients.
  • 40% of any composition should be empty space.
  • When in doubt, remove.

North star metrics

  • Daily active users (intentional retrieval: shortcut, search, timeline scrub) — not app launches.
  • Activation rate — % of app openers who perform an intentional action.
  • D7 retention — do they come back?

What we believe

  • Humans spend 8+ hours/day on screens. Digital work is bottlenecked by human attention. AI needs full context to truly automate.
  • Vision is the universal interface — it's the most powerful sense of most animals.
  • We believe humans should be augmented by AI to focus on more profound exploration instead of repetitive digital labor.
  • Positive sum games. Give first. Optimize for the long term.
  • Radical transparency. No ego. Idea meritocracy. Constructive disagreement.
  • Truth over comfort. Bold over safe. Ship over plan.