cynarAI 6c7eed7580 Initial clean evOo-local from f00577b

Fresh history cut from cynarAI/evoo antigravity-handoff/project-timeline
(Voice First + project timeline). Brand docs/titles as evOo; keep @rakazo/*
packages as tech debt. Excludes node_modules, caches, secrets, voice blobs.

2026-09-20 17:15:16 +02:00

3.7 KiB

Raw Blame History

Persona & crew selection — personalization bar

evOo must adapt to each user more individually than any other app. Persona/crew fit is lifestyle-agnostic: office worker with hobbies, club/scene person, hardcore programming nerd — the product does not care. They get what they want in the simplest way.

Accessibility of ambition (skill floor / ceiling)

It must not matter whether the user is a power user. One product; radically different crews and Unterhaltungen.

Example need	Crew shape	Entry	Depth
Thermomix help (mom)	Home/kitchen specialist	Grandma-simple	Excellent in that domain
Hard mathematical calculation	Quant / math specialist	Plain ask, no LaTeX required	Expert-grade
Software development	Engineer / architect Oos	Simple “build this”	Expert-grade

Skill floor = grandma-simple — no power-user literacy required to start or win.
Skill ceiling = expert-grade per specialty — not one bot that is mediocre at everything.

Onboarding and shell copy stay on the floor; spawned specialists and thread depth carry the ceiling. Never conflate “simple UI” with “dumb agent.”

Emergent ensemble

More individualists > one generalist. Each Oo is a narrow expert; value emerges in Gruppenunterhaltungen when different lanes combine (handoffs, peer replies, parallel work). Never design copy or flows that suggest one Oo could replace the crew.

Ties: no Wunderwuzzi · specialty depth · group fun friction · docs/GROUP-CONVERSATIONS.md.

Relationship over time

Principle	Persona/memory
Always on	Steady availability in copy — not nagging
Compounding	MEMORY + threads deepen with use; onboarding seeds only honest signals
Never creepy	No inferred lifestyle; consented harvest only
Not yes-men	Pushback in-lane; voice copy allows disagreement (`docs/OO-RELATIONSHIP.md`)

Companion tone (never one personality for everyone)

Mode	When
Geek-peer / Fachsimpeln	User wants depth — match enthusiasm, stay in lane, no lecture
Quiet-doer	User wants outcomes — execute, hand off, never act like we know better

Both modes forbid condescending Wunderwuzzi behavior. Specialists defer out-of-lane work to other Oos.

Inference → spawn/select (not assign)

Onboarding harvest builds hints in user MEMORY.md (user-help-vibe.ts):

Observed signals — computer attached, self-hosted model, locale, focus choice, connectors (when known)
Suggested specialist lanes — archetypes from bot-persona.ts, not lifestyle buckets
Interaction style — unknown until focus choice or conversation; then geek-peer or quiet-doer

Focus choice (onboarding)	Typical lanes	Interaction lean
Day-to-day work	inbox-work, chief-operator, curator	quiet-doer
Inbox & email	inbox-work, inbox-life, chief-operator	quiet-doer
Research & writing	curator, builder, engineer	geek-peer
Specialist crew	chief-operator, specialist (+ add as needed)	learn first

TODO: Use hints to propose Unterhaltungen, spawn templates, and project folders — never a single default crew.

Archetype detection (existing)

Runtime persona comes from bot name / title / description → resolvePersonaArchetype() in apps/web/src/lib/bot-persona.ts. Templates under templates/ are starting points; user fit beats template defaults.

When writing template copy or voice lines: narrow scope, peer tone option, no omniscient helper voice. Default utterances assume floor literacy; specialist instructions and tools assume ceiling competence in-lane.

3.7 KiB Raw Blame History