North-star architecture · directional mockups

Conversational Copilot + Adaptive Brand Memory

Chat becomes the front door; the existing screens become the live canvas it drives. You type a requirement, the copilot understands it, offers options, confirms a plan, and — only on your approval — applies it. Corrections and preferences become scoped, reversible learnings that get injected into future work and graduate into the profile / style guide over time. And it's portable: paste plans in from ChatGPT/Claude, export a hand-off package, bring the result back.

The 6 locked decisions

Front door + canvas

Chat is primary; screens are the live canvas it drives

Approval-gated, inline

Understand → options → plan → approve → execute; never silent

Hybrid learning

Scoped reversible rules that graduate into structured edits

Universal interop

Paste-in · export package · bring-back + keep connectors

Reuse the engine

Typed capabilities wrap existing routes; heavy gen unchanged

Brand Copilot first

Shared substrate → Brand Copilot → Creation Copilot

Walk the suite

1 · Front door

Talk to your brand — quick-start intents, recent conversations, cold-start

chat-as-home

2 · Brand fix — the full loop centerpiece

Understand → options → form → plan → approve → apply → remember, with a live profile canvas

the loop

3 · Memory & learnings

Scoped rules (item/surface/brand), status, reinforcement, and a graduation proposal

4 · Style-guide ladder

Refine an image → the 4-rung design-change ladder → the style guide stays living

creation taste

5 · Interop bridge

Paste a ChatGPT/Claude plan in · export a hand-off package · bring the result back

portable

Directional mockups for the approved north-star design (docs/superpowers/specs/2026-07-06-conversational-copilot-adaptive-memory-design.md). Demo brand: Fidelis Global (enterprise relocation). Next: iterate on these → per-surface detailed specs → build.