VIBEnet is the audible awareness layer between humans, agents, and devices. This lab publishes the twelve proofs-of-concept that demonstrate that claim end to end — from a single agent heard through a browser, to a cross-device mesh that renders one signal into audio, voice, AR, and cognitive gating. Each rung below has a status. Most are planned or in progress. 3 are shipped.
Direct answer
What is the VIBEnet research ladder?
Twelve proofs-of-concept that extend one Signal Contract from a single agent heard in a browser to a cross-device mesh of audio, voice, AR, and cognitive gating. Each rung carries an honest status — shipped, in progress, planned, or research — and SERPRadio sits outside the ladder as the production proof.
LIVE ANCHOR
Pulse Trace: Hear the agent run.
The working proof lives on the canonical demo so the lab can stay a readable notebook. That proof now rides the same scored reference clock from hero through trace. Start there, then come back here for the twelve-rung ladder that extends the same contract across renderers, devices, and domain adapters.
shippedRunning in production. Linked demo available.
in-progressActively building. Partial demo or prototype.
plannedScoped, not yet started.
researchStill forming. Platform, ethical, or technical questions open.
01
shipped
VIBEnet Pulse
I can hear what the agent is doing.
nominaladvisorywarninghandoffrecovery
Research detail
A single autonomous agent runs a task. Every meaningful execution event: planning, tool call, retrieval, warning, handoff, completion. Each one emits a Signal Contract event, which VIBEnet converts into a live ambient audio layer. The visitor listens to the agent work.
Proves agent awareness by ear. A cold visitor should be able to close their eyes for twenty seconds and describe the arc of the task afterward. This rung is the primitive. Every higher rung builds on it.
browser·tone.js·core product
02
shipped
Codex Lifecycle Replay
See attention without seeing source content.
Research detail
A deterministic Codex app-server fixture passes through an independent VIBEnet adapter. The replay exposes unavailable, idle, running, approval, error, recovery, completion, and settled-idle states as validated Signal Contract events without carrying source content.
Proves lifecycle state can become a privacy-preserving attention field. HMAC references replace raw identifiers, error outranks ordinary activity, recovery emits once, and completion settles after a deterministic hold. The proof is a documented replay, not passive live fleet observation.
Watch domain meaning become authorized perception.
Research detail
A browser-local Domain Adapter workbench. Bring a foreign JSON/JSONL stream, accept deterministic mapping rules, and watch each source event either become a validated Signal Contract object—or stop at the adapter with no invented sonic state. A/B profiles make a bad taxonomy audibly flatten critical distinctions.
Proves: observed ≠ understood ≠ authorized to render. Machine assist may propose rules; only ratified rules emit. Contour answers how meaning moves through time; Foundry answers where meaning came from; the Perceptual Mixer (downstream) decides what deserves foreground attention without rewriting those contracts.
The human talks, the agent answers, the system sings underneath.
advisorywarninghandoff
Research detail
The human speaks through phone or glasses microphone: what are you doing, why did tension rise, pause, summarize the last sixty seconds, escalate only if it gets worse. The agent responds in voice. Underneath both voices, ambient VIBEnet audio continues, keeping the human oriented to system state across the whole conversation.
Proves that voice is the explicit channel and VIBEnet is the ambient channel. Those are different jobs. Speech says things; ambient audio maintains awareness.
browser·websocket·voice agent
05
planned
Ray-Ban Audio Companion
Glasses become agent-awareness earbuds.
nominaladvisorywarninghandoff
Research detail
Phone runs a companion app or PWA. Ray-Ban Meta glasses act as open-ear audio output and microphone input over Bluetooth. The human walks, cooks, commutes, codes, or sits at the piano while agent state plays continuously in open air. A route anomaly triggers a motif shift. The human asks what changed. The agent answers.
Proves that post-screen agent awareness works with consumer glasses today, before any display SDK access. This is likely the fastest glasses demo available.
ray-ban meta·ray-ban display·oakley meta·phone bluetooth
06
research
Meta POV Agent
The glasses camera gives the agent eyes.
advisorywarninghandoff
Research detail
Use the Meta Wearables Device Access Toolkit to let a mobile companion app receive camera context from compatible AI glasses, then forward that context to a VIBEnet agent. The agent sees what the wearer sees — a laptop dashboard, an airport departures board, a printed meeting agenda, a piano session — and converts visible complexity into sonic summary.
Proves VIBEnet can become a wearable perception layer, not only a website. Safe first demos target objects, documents, dashboards, and context. Face recognition and bystander identification are explicitly excluded.
meta WDAT·camera·mobile companion
07
planned
Snap Spectacles VET Lens
Agent state becomes AR atmosphere.
nominaladvisorywarningrecoveryhandoff
Research detail
A Lens Studio experience in which VET state renders as spatial AR above the wearer's workspace. Low tension shows as a soft stable aura. Higher energy speeds particle motion. Warning pulses amber. Completion expands a ring. Agent handoffs travel as arcs between nodes. In multi-agent rooms, each agent has a floating sonic and visual identity.
Proves VIBEnet renders across audio and spatial visuals together. First demo: one floating agent orb above a laptop, changing color, motion, and sound with live agent state.
snap spectacles·lens studio·snap cloud
08
research
Crown Cognitive Gate
The agent learns when not to interrupt.
advisorycriticalhandoff
Research detail
Use the Neurosity Crown's focus and calm probability metrics as a soft gate on VIBEnet's interruption density. High focus reduces audio intensity. Low focus permits richer explanatory narration. Low calm routes non-urgent alerts away. Critical agent events break through regardless, with a distinct warning motif.
Proves that human-agent communication can respect the human nervous system. Framed as cognitive availability sensing, not mind reading. This is the politeness primitive — agents learning when, how, and how loudly to enter human awareness.
neurosity crown·eeg·cognitive adaptive interface
09
research
Kinesis Agent Control
Thought gesture as a command layer.
handoff
Research detail
Use Neurosity Kinesis for two or three trained motor-imagery commands. Low-stakes actions only at first: pause or resume sonification, request agent explanation, mark the current moment as important, lower audio density. Every Kinesis action requires audible confirmation before execution.
Proves human-agent communication does not have to route only through keyboard, mouse, voice, or touchscreen. Sequenced after Cognitive Gate because safer gating must be proven before command execution.
neurosity crown·kinesis·motor imagery
10
in-progress
Human-Origin Hybrid Composer
One piano note becomes an adaptive agent language.
nominaladvisoryrecovery
Research detail
A human plays a motif at the piano — one note, a two-note interval, a phrase, or an improvisation. The system captures MIDI, velocity, timing, sustain, phrasing, and tags the gesture with VET coordinates. AI augmentation then harmonizes, orchestrates, stretches, inverts, layers, and spatializes. The same human-origin motif appears as browser piano, Ray-Ban earcon, Spectacles particle pulse, agent trace cue, Crown-adaptive variation, and live route signal in SERPRadio.
Proves the system is neither human-only music nor generic AI music. It is human-origin, AI-augmented sonic intelligence. The authorship nuance is the moat.
A mesh is not impressive because five devices can receive JSON. The hard problem is that five devices—and one person—remain coherently aware of thousands of concurrent sources. PerceptualMixer01 ships the bandwidth layer first: governed foreground vs ambient on a 64-agent fixture, with suppression receipts and RAW vs GOVERNED A/B. Devices come after attention is honest.
Proves the fabric can share one nervous system without deleting evidence. Full multi-device keynote remains planned; the Mixer is the executable prerequisite. Fabric Reference Run 01 demonstrates the full stack end-to-end: Source → Foundry → Signal Contract → BeatSync → Contour → Mixer → Renderers → Receipts.
Long-running generation work surfaces as audible state.
advisorywarningrecoveryhandoff
Research detail
Route Higgsfield MCP runs — Sora, Veo, Kling, and the hosted catalog — through the same event bridge as planning and retrieval. A run start, model swap, preview ready, or stall becomes a Signal Contract stream you can hear before the preview renders.
Proves generation composition is not a separate product lane. It is another adapter path on the same contract and clock.
SERPRadio is the first live Domain Adapter — route intelligence, materialized state, and audio previews already in production. It is not rung twelve. It is the production proof that the contract works on a live domain surface.
Every rung speaks the same Signal Contract. One JSON object carries an entity, an event, a public semantic channel, and six flat renderer-facing state fields. Each device adapter renders that signal differently: a browser shows a trace, glasses play an earcon, Spectacles draw an aura, the Crown gates interruption. The contract is the shared surface. Let adapters proliferate while protected source material and rendering decisions stay governed.
One object carries entity, event, channel, valence, energy, tension, intensity, hue, and pulse. Read the schema JSON for the canonical object and the governance page for the public build path.
One human-origin motif. One live agent. One signal contract. Many devices.
LISTENING SESSION
Agents work. You hear it.
Pulse teaches the language. The lab shows the ladder. Join the list for protocol updates, demo drops, and the first listening-session invites.