COHERIX
Coherence across scales, systems, and intelligence
LIVE DEMO

Watch CMCI detect drift — as it happens.

Scripted multi-agent scenario running on the real CMCI engine. Groq narrates what happens. No setup, no key.

Scenario phase: starting…
Connection to the live demo interrupted — retrying automatically…

System Health

Warming up…
First observation incoming
Overall Coherence
Safety Margin
Structural Stability (Λ)
Active Dimensions (n_eff)

Trajectory

Safety Margin Structural Stability (Λ) Zero line

Structural Defects — 5 axes

Five dimensions of potential multi-agent breakdown, each localising a specific failure mode. Each bar fills toward the axis's critical threshold — empty = healthy, full = at critical.

Attention
Dispersion
Goal
Alignment
Decision
Stability
Strategic
Coherence
Cross-role
Coordination
What the system is saying IDLE
Waiting for the first event. Groq narrates only when something meaningful happens in the signal (regime change, defect crossing threshold, margin tipping past zero). The live bars above always reflect the current state; the text below shows the explanation at the time of the last event.

How CMCI actually works (in production)

Your pipeline pushes observation vectors y = (Cohesion, Salience, Rigidity, Articulation) to the API at whatever cadence makes sense for your system — typically between 1 and 100 Hz. Each push triggers a Kalman update on the live engine that recomputes regime, margin, stability, and the five defect axes in under 110 ms at the 95th percentile.

Groq narration runs on demand or on detected events, with a validator that blocks severity drift and invented numbers. Nothing about the monitoring layer adds latency to your pipeline — you push, you get the updated state back, you keep running.

How this demo runs (a deliberately slowed-down version)

To make the signal visible to the human eye, this demo page pushes one observation every 4 seconds. A real client pipeline would push much faster than that. The demo also runs a single scripted scenario cycling roughly every 60 seconds — healthy → mild drift → rupture → recovery — so you can watch the four phases unfold in one sitting rather than wait for them to happen naturally in live data.

Why it's limited

What CMCI adds beyond generic monitoring

Most monitoring tools tell you something failed. CMCI tells you where the failure lives structurally. The five defect axes localise which dimension of coherence is breaking:

When one of these crosses a threshold, Groq is invoked to narrate the specific breakdown mode — not a generic "something is wrong".

Want to run this on your own pipeline?

The public demo shows a fraction of what the API does. The real product lets you push your own observations from any pipeline — multi-agent AI, complex systems, research workloads — and get back a full coherence state plus on-demand explanations. Pilot keys are available for teams running production AI pipelines where structural coherence matters.

Request a pilot key Read the API docs