Reference · ESG agent

How the Horizon ESG agent works.

A plain explanation of what the agent is, what it measures, where the numbers go, and what it deliberately doesn't touch. No pitch, no pricing — those live on other pages. This one is here so the people doing the technical, security, and sustainability reviews can read what the thing actually does.

Collectors

5

scoped to scope 2 only

Conversion factor

0.207kgCO₂e/kWh

desnz · ghg dual-reporting

Reporting

SECR-shaped

24h · 7d · 30d · annualised

01 · What it is

What it is.

The Horizon ESG agent is a small program that runs on a server, desktop, or appliance and reports how much electricity that machine is drawing. It posts the readings to the Horizon Platform once a minute over HTTPS. The portal applies a published grid-emission factor, joins the reading to your electricity tariff, and stores the result against the company the agent belongs to.

It is intentionally narrow. It is not a monitoring agent, an EDR agent, an asset-management agent, or an inventory agent. It runs five collectors, all of them related to energy and heat. The rest of the Horizon Platform collector set is not started on an ESG-mode agent — the surfaces that would emit security, network, or software-inventory telemetry are never loaded into memory, and the server-side ingest rejects any payload that arrives on them anyway as defence-in-depth.

The shape of the output, methodology, and provenance discussion lives on Measurable AI. This page is about the agent that produces the readings underneath it.

02 · What it measures

Five collectors. Energy, heat, and apportionment.

Five collectors run on an ESG-mode agent. Each is scoped to a question about energy, heat, or how those numbers should be apportioned across the workloads sharing the host.

Power. Whole-host electricity draw in watts. Read from primary counters where the platform exposes them — RAPL on x86 Linux, IPMI on baseboard management controllers, NVIDIA / AMD GPU counters where present, powermetrics on Apple Silicon, LibreHardwareMonitor on Windows. Where no primary counter is reachable, a calibrated CPU-time × TDP estimate is used and the reading is labelled estimated so the figure travels with its method.
Sensors. Thermal readings (CPU, GPU, NVMe, motherboard) and fan tachometer speeds. Useful for the heat-dissipation and cooling-load calculations in the report, and for spotting an endpoint whose draw rose because its cooling failed rather than because its workload grew.
Hardware. CPU model, core count, RAM, motherboard / chassis identifier — read once a day. Lets the portal pick the right TDP profile for the host and recognise hardware refreshes when they happen.
Docker. Container inventory and per-container CPU%. Used by the portal to attribute the host's power draw across the containers running on it, proportionally to their measured CPU time. The allocation method is stored with the figure.
Health. CPU%, memory%, disk%, uptime. The basic liveness signal — without it the portal cannot tell "the agent stopped" from "the host is genuinely drawing zero watts".

03 · What you get back

Auditor-friendly numbers.

The agent itself emits readings. The numbers an operator or auditor reads live in the Horizon Platform's ESG Reporting tab, which the agent populates:

Energy. kWh on 24-hour, 7-day, 30-day, and annualised windows — at endpoint, location, and fleet level. Annualised figures are explicitly labelled as projections.
Scope 2 emissions. kg CO₂e, with GHG Protocol dual reporting: location-based using the current DESNZ UK Government GHG Conversion Factor (0.207 kg CO₂e / kWh at the time of writing), and market-based when you supply a supplier-specific factor and a REGO certificate reference.
Electricity cost. £ on the same windows as the energy figure, using your contracted tariff. The unit-cost source is stamped on the report.
Heat dissipation. IT waste heat, cooling overhead at a default PUE of 1.5, and the resulting facility load in BTU/hr. Useful when the conversation moves to HVAC sizing or a heat-recovery plan.
SECR intensity ratio. kg CO₂e per £m revenue, if you've entered an annual turnover figure. The shape auditors of quoted entities expect.
Per-host thermal maximums. 24-hour peaks across CPU, GPU and NVMe. Worth knowing before a capacity conversation.
CSV export. One row per endpoint per measurement window, with timestamps, workload IDs, factor sources, method labels and unit-cost sources. The trail an auditor follows.

What it never touches

The ESG-mode agent runs a strict subset of the collector set. The surfaces below are not loaded — there is no code path on an ESG-mode host that can emit them, and the server-side ingest discards them if they ever did arrive.

No network packet captures, port scans, or any active network probing.
No software inventory beyond the OS / kernel string the host reports about itself.
No cybersecurity telemetry, antivirus state, or firewall configuration.
No file contents, file paths, or application data.
No system log content, journald entries, or audit logs.
No process command lines or environment variables.
No user accounts, identifiers, or access events.

This is a deliberate scope choice. An agent that runs everywhere has to be auditable everywhere, and the simplest way to make a privacy / security review go quickly is to have less for it to review.

Where the data lives

Hosting. UK-hosted Postgres, single-tenant logical separation per company. Telemetry does not leave the UK.
Direction of traffic. Outbound-only HTTPS from the host to the portal. The host is never reachable from the portal side — there is no inbound shell, no listening port opened by the agent, no return channel.
Retention. 90-day rolling window by default. Fixed-period retention is configurable on request and is enforced by automated deletion, not by policy.
Update channel. The agent pulls signed tarballs and checks a recorded SHA-256 before applying. A failed update rolls back automatically; a tampered tarball never installs.
Uninstall. One-line removal that leaves no residual state on the host beyond the install-log entry. After uninstall the agent goes offline in the portal within five minutes and is auto-archived after thirty days.

How it deploys

Deployment is deliberately short — most of the work in an ESG programme is in the methodology and the reporting, not in getting an agent to run.

Generate a token in the portal. Pick the company, tick "ESG mode", and copy the install command.
Run the install command on each host. A single line on Windows (PowerShell as admin), macOS, or Linux. The installer drops the agent into a known path, registers it as a service or LaunchDaemon, and exits.
Wait a minute. The agent registers itself with the portal on first run, and the first power reading lands within sixty seconds. The ESG Reporting tab begins populating once a handful of readings exist.

The same install command runs across Windows, macOS, Linux, Proxmox hosts, and pfSense edge devices. The collector set adapts to what the platform exposes; the reporting shape does not.

What we won't claim

An honest list. We'd rather name the gap than pretend it's solved.

Estimated, not metered. Power figures derive from primary counters where the platform exposes them and from calibrated CPU-time × TDP estimates as a fallback. Per-container attribution is proportional. The method label travels with every figure.
Auditor-friendly, not third-party verified. The agent produces primary data with a disclosed methodology — the substrate that reasonable assurance can rest on. Attestation per ISO 14064-3 requires an accredited verifier (LRQA, BV, DNV); the judgement still belongs to the auditor.
Scope 2 only. Purchased electricity for IT infrastructure. Scope 1 (direct fuel, refrigerants) and Scope 3 (everything else) are out of scope and we don't pretend to measure them.
No offsets, no "carbon neutral" framing. We measure and report. What you do with the figures is yours, and we don't sell anything that would make the answer cleaner than the measurement.

Measurable AI

The methodology pillar: how source telemetry becomes a defensible figure, the disclosed-factors table, and the accuracy bands attached to each method.

Read Measurable AI

Sustainable infrastructure

Where the carbon actually sits, what heat reuse looks like in practice, and why sovereignty and sustainability are the same conversation.

Read Sustainable infrastructure

ESG hub

The three pillars together — measurement, infrastructure, governance — and how they fit into a disclosure that survives a hostile review.

Back to the ESG hub

Next step

See it in the platform.

The ESG agent is one of three modes shipped by the Horizon Platform. See where the kWh, kg CO₂e, and £ figures land in the platform's ESG reporting tab.

See the platformSee the platform Bring a workload

No deckNo sales pressureWe'll tell you to wait if you should

Or email a workload to hello@althorizon.co.uk — one-page model back in 48h.