Better CMDK

From user intent
to deployed product.

The cmd-K that ships changes. A user asks for something in the palette; an agent factory builds it; a preview comes back to that same user, in the same app, for sign-off.

Open the live dashboard ↗

How it works

The loop

Five moves from a wish to a shipped change. No tab-switching, no ticket triage, no handoff where the request loses its context.

  1. 01

    Hit cmd-K in any app

    The palette is embedded in the product itself. Wherever the user is, the command surface is one keystroke away.

  2. 02

    Describe the change

    The user says what they want in plain language. Their last actions and a recording of the session attach automatically as evidence — the request arrives with proof, not a vague repro.

  3. 03

    The agent factory takes it

    Agents plan the change, implement it, and QA it against the captured evidence. Humans review where it matters; the factory does the rest.

  4. 04

    A preview comes back in-app

    The user who asked gets a preview deployment, inside the app they asked from, and signs off on the exact thing they requested.

  5. 05

    Applied per surface

    Sign-off promotes the change surface by surface. Every deployment traces back to the intents that caused it — the audit trail is the product.

Why one unit

Stop stitching point solutions

Today the intent-to-deployment path is spread across four tools that don't share an object model. Better CMDK wraps them into one deployable unit whose core object is a user's change request — and the deployed answer to it.

Session capture PostHog-style replays
Evidence attached to every request, automatically
Error & feedback capture crash reports, NPS widgets
Intents in the same stream as everything else
Ticketing Jira, Linear, a spreadsheet
Change requests that carry their own context and state
Deployment management scripts, dashboards, prayers
Previews and applies, per surface, traced to intents

One deployable unit. One event log. One object that goes in as a wish and comes out deployed.

Live demo

Try it live

The station dashboard below is the real control plane — comments, crystallizations, tickets, runs, deployments. The palette is embedded in it: press K (or CtrlK) once you're there.

live station 34.187.233.150.sslip.io/dashboard ↗ The hosted control plane. The palette is already embedded — cmd-K works there.

This site is itself a Master CMDK surface: the real SDK loads from the station and registers this page as better-cmdk-site. Anonymous requests run in demo mode.

Principles

What the system believes

01

Event sourcing

Capture everything. The event log is the substrate — for insight, for demo videos, for demo sequences, for the QA evidence agents check their work against. Nothing is derived that can't be replayed.

02

Intent capture

Catalog what people wanted while they were using the app — users, engineers, support alike. Wishes are data. The backlog writes itself at the moment of friction, not in a retro three weeks later.

03

Room for the next principle. The loop is young.