Verdify launch video: the greenhouse, AI planning loop, ESP32 safety boundary, Slack Ops, and public telemetry story.

The Verdify lab is the public proof point: a real greenhouse in Longmont, Colorado where AI planning, edge control, telemetry, and operations all have to work under physical constraints.

This site publishes the proof: plans, telemetry, failures, costs, lessons, and scorecards so readers can check whether the system works instead of taking the claim on trust.

Right Now

The panels below are live Grafana trends from the public read-only telemetry path. They show 72 hours of observed greenhouse climate plus the forecast window the AI planning agent is planning against.

Temperature loop

Greenhouse temperature against the active control band. The green band is the firmware target envelope stitched from recent controller setpoints into the forecast window. Read the graph as a control trace: greenhouse temperature, outdoor pressure, forecast pressure, relay state, and solar load all in one place.

Moisture loop

Vapor pressure deficit shows plant drying pressure. VPD is how hard the air is pulling water from leaves. High VPD is dry-air stress; low VPD means the greenhouse is too humid for healthy transpiration. The panel pairs the greenhouse trace with forecast and outdoor context so the moisture posture is visible before relays fire.

Lighting loop

Solar lux, grow-light threshold, and confirmed light state. The yellow area is observed and forecast solar context. The green band is the active grow-light threshold window, and the blue state blocks show when the main grow-light circuit is actually on.

Live Greenhouse Cameras

Live public snapshots show the greenhouse behind the telemetry: crop benches, hydroponic channels, service hardware, lighting, and day/night state.

Latest public snapshot from greenhouse camera 1
Greenhouse camera 1. Snapshot refreshes in-browser every 30 seconds.
Latest public snapshot from greenhouse camera 2
Greenhouse camera 2. Snapshot refreshes in-browser every 30 seconds.

What To Look At First

What is the proof pattern?Verdify Greenhouse Case Study

The polished case study: AI may plan, but authority must be bounded, validated, measured, and operated.

Does it work?Baseline vs AI Planning Agent

The April 22-25 planner-offline window is compared with the following planner-online window.

Is it safe?Safety Architecture

The canonical safety page owns the relay-boundary explanation.

Can I inspect the data?Evidence map

Start there for the live pages, generated archives, APIs, and sample exports.

Why This Is Worth Checking

Real physical stakes

This is a high-elevation Boulder County greenhouse: cold nights, intense sun, dry spring air, summer heat, winter snow, and fast shoulder-season swings.

Planning memory

AI Greenhouse Control explains what the AI planning agent reads before it writes a plan.

Public evidence

Evidence pages are the routing layer for live state, scorecards, archives, and sample data.

Human operations

Operations turns plans, deviations, reminders, and operator tasks into human-readable messages without making Slack a safety layer.

Honest resource accounting

Resource Use separates solar timing, electricity, gas, and water so the environmental story does not hide the remaining utility bill.

Claim Boundary

ClaimAI-assisted tactics can be audited against physical outcomes.

The proof layer is climate, stress, cost, water, failures, lessons, and plan outcomes.

Not claimed yetYield, profit, and full autonomy are roadmap claims.

Those need more harvest records, crop-stage normalization, and comparable baselines before they belong here.

Further Reading