Gen-2 harness — one engine, many surfaces, built for others too
The orb moves to a persistent left dock; the main panel scrolls and swaps content (grid, conversation, dashboard) without the orb ever leaving. Multimedia cards are equal citizens. And the whole engine repackages into Mentor for other people. Toggle the states below.
▲
PORTFOLIOSTACKHARNESS
hold to talk
APPSCONTENT≤30d⊞ grid◆ GOD
Byron Biography
updated 2d
Sensorium
updated 5d
▶ VIDEO · 1:48
▶
Byron — Nocturne (piano)
performance · rel 2026
AgentLand
updated 9d
Mentor
updated 3d
AI Benchmarks
updated 12d
🔊 AUDIO
RAI Report
updated 1d
Snowscape
Smithsonian '26
Morning. EP 15 rendered clean. Want me to pull the piano card into your reel for the demo?
Gia · also in Telegram
yes, and add Benchmarks + Biography to context
Done — 3 cards in context. Ask me anything about them, or say "make a 2-minute intro."
◆ IN CONTEXT
Piano · perf
Benchmarks
Biography
What this answers: the orb/action becomes a persistent dock, not a toggling button — it's always there, the panel is what changes. Going to grid is just the panel's content, orb untouched. Conversation keeps the same dock + adds a slim context rail (right) so portfolio and chat are alive at once — multiple surfaces, one frame. Chips/filters ride the panel top. Gestures/keys: the deck's arrows/pinch stay inside the panel content; the orb owns cross-surface travel (hold=talk); no collision because they operate at different layers.
Multimedia cards = equal citizens
You asked to see it: the piano video card sits in the same grid at the same weight as an app — full-bleed frame, ▶, duration badge, a live waveform, "performance" provenance. Same for an audio card (🔊). The card species now spans illustration · photo · video · audio · generative — the deck doesn't care what's inside, only that it has a face and a provenance. That's what makes "employ me — here's me playing piano" a first-class tile you can drop in a Stack and send.
▸ REPACKAGED FOR MENTOR — the same engine, scoped to someone else
A mentee registers → taps "Generate my profile" → the engine builds their portfolio from their context only (never yours). "Generate my Signal" → an action-oriented podcast about their work. Tiered:
Basic 2-min podcast · starter profile · a few cards
Harness (top) their own personal-AI surface — the big claim
Naming knot you flagged: "Signal" is doing three jobs (public podcast · your private data view · the harness itself). I'd split: The Signal = the public podcast (keep), Brief or Situation = your private data surface, and the whole thing = the Harness (or a new coined name — worth a proper domain pass, not the sprays that came back all-taken).
⚖ WHAT YOU DIDN'T ASK — a pass against your actual goal (next-gen, extensible, anti-portal, for-others)
1 · Tenancy isolation is the whole ballgameThe moment one engine serves you AND a mentee, the #1 risk is context bleed — their "generate profile" must be provably unable to reach your data or another mentee's. That's a multi-tenant boundary designed in from line one, not added later. It's also your RAI credibility: you, of all people, cannot ship a tool that leaks across users.
2 · Whose artifact is it?If you generate someone's portfolio + podcast, you owe them a clear answer on ownership, portability, and retention before you ship — can they export it, does it die if they leave, what did you keep. A mentor program lives or dies on that trust contract.
3 · One orb ≠ a dashboardHonest tension: the orb travels a linear stack beautifully, but a true multi-panel dashboard (several live surfaces at once) is a different spatial model. Don't force one metaphor onto both — phone = orb-travel, desktop = dockable panels. Same engine, two shells.
4 · The cost meter per personGenerative portfolio + podcast per mentee is real spend. The tiers ARE the throttle, but each tenant needs a budget guard (the voice cost-guard pattern) or a viral moment bankrupts it.
5 · The moat, said plainly"Personal AI" is the most crowded claim in tech. Yours is different for one reason: it's governed and owns-the-canon — provenance on every card, disclosure on every voice, context you can see and control. Lead the pitch with that, not the chat. It's the one thing an RAI leader can say that a chatbot company can't.
6 · Multimedia changes the pipelineVideo/audio cards are great — but they're a new asset class (hosting, transcode, thumbnail, rights) the art/motion pipeline never handled. Small, real, and worth a dedicated "media card" ingest path rather than bending the image pipeline.
💡 The reframe worth keeping: you're not building a portfolio with extra modes — you're building a personal-AI operating surface whose first tenant is you and whose product is "give anyone a governed version of this." Portfolio, Command, Signal, Mentor aren't features — they're the same engine pointed at different contexts. That framing is the anti-portal architecture you keep reaching for, and it's genuinely novel.
Status: concept + clickable mockup only, nothing built. Preserved from prior asks: chips/filters, Stack + share, provenance, the orb, theme fidelity. Target-state note held for Mentor/Command/Signal integration.