Run AI without sending data to someone else's cloud
Self-hosted deployment on Docker, PostgreSQL, and Redis. Agent workloads run inside your perimeter. Your data, your logs, your retention policy.
Product · Horizon Platform
The AI management platform that runs where your data lives.
Horizon Platform gives operations and governance teams a single surface for AI-assisted workflows, fleet visibility, and policy control. Telemetry comes from one cross-platform agent that ships in three modes — full Horizon, Cyber, and ESG — so each host runs only the collectors it needs.
Architecture
1 platform · 3 modes
full · cyber · esg · server-enforced
Surfaces
Cross-platform
windows · macos · linux · proxmox · pfsense
Trust posture
Self-hosted
outbound-only https · uk postgres · 90-day retention
One surface for AI-assisted operations
One installer, three modes. Pick the surface each host needs.
The Horizon agent ships in three modes. Same installer, same signed update channel, same outbound-only path to the portal — the collector set is rotated to fit the question the host is answering. You pick the mode when you generate the install token; the server enforces it on every check-in. Hosts can be moved between modes by issuing a new token.
Horizon (full)
Fifteen collectors covering health, hardware, containers, networking, filesystem layout, software inventory, security posture, electricity draw, thermal sensors, AI-model presence, Proxmox & PBS hypervisor surfaces, severity-filtered system logs, Hermes AI project context, and Cyber Essentials Plus evidence. The mode for fleets you operate end-to-end. Server-side mode enforcement on every check-in.
Read the Horizon agent referenceCyber
Eight collectors scoped to Cyber Essentials Plus A4–A8 evidence (firewalls, secure configuration, updates, user access, malware) and CIS-style topology scoring. Software inventory cross-checked against endoflife.date for fifteen security-critical packages. Asset inventory pre-populates CE+ A2; export is a branded PDF in IASME question-order. Multi-tenant super-admin support. Observability only — no remote-control surfaces.
Read the Cyber agent referenceESG
Five collectors scoped to energy, heat, and apportionment metadata. Energy on 24-hour, 7-day, 30-day, and annualised windows. Scope 2 emissions via GHG Protocol dual reporting — location-based on the current DESNZ factor (0.207 kg CO2e/kWh), market-based with your REGO certificate reference. SECR intensity (kg CO2e per £m revenue) if turnover supplied. Scope 2 only — not Scope 1 or 3.
Read the ESG agent referenceAll three modes share the same UK-hosted Postgres, outbound-only HTTPS, 90-day default retention, signed update channel with SHA-256 verification, and one-line uninstall. The differences are in scope, not in trust posture.
From source telemetry to defensible Scope 2.
Most technology ESG reporting breaks at one of two points: data quality, or access to it. If your numbers can't be traced from source telemetry to reported figures, they're hard to defend under real scrutiny.
Horizon Platform is instrumented differently. The included ESG agent runs across the heterogeneous fleet most operators actually have on the floor — Windows, macOS, Linux, Proxmox, pfSense — and joins primary kilowatt readings to your grid-emission factor and your contracted electricity rate. You get measured energy, Scope 2, and cost views, with the chain of custody every link of the way.
For AI workloads specifically, the same chain runs end-to-end: the token a user generated, the GPU-second that produced it, the kilowatt the rack drew, the grid factor your jurisdiction publishes, and the CO₂e you put on the sustainability report. Cloud AI shows you the bookends; Horizon Platform shows you everything in between.
80.1 kWh
Measured across 10 endpoints — Windows, macOS, Linux, Proxmox, pfSense — on our own infrastructure.
16.6 kg CO₂e
UK grid factor; full provenance in the disclosed-factors table.
£24.03
Ofgem UK 2024 unit rate. Replace with your own contracted rate at deployment.
From our own kit, not a customer reference. We dogfood the ESG agent before we deploy it on yours. See the ESG agent product page →
Cross-platform telemetry, one agent
A single ESG agent runs across Windows, macOS, Linux, Proxmox, and pfSense, with primary counters where the platform exposes them and labelled estimation where it doesn't. Other hypervisors and operating systems are on the roadmap. See the ESG agent →
See every workflow's energy cost, per run
Each completion, summarisation, or agent run is joined to the GPU time, CPU time, and kWh that produced it. Power draw and token throughput appear on the same dashboard, timestamped per workflow. The same view your engineering team uses to tune cost is the view your sustainability team uses to report.
Translate kWh into the numbers ESG frameworks actually want
Horizon Platform applies your region's grid-emission factor (or your own contractual one) to convert kWh into CO₂e, and your contracted electricity rate to convert kWh into £. Outputs are exportable as a CSV trail with timestamps, workload IDs, factor sources, and method-accuracy badges. The kind of thing an auditor can trace.
Compare self-hosted against cloud, honestly
For every AI workload, the Portal shows what the same tokens would have cost (in money and in carbon) on the major cloud providers, using published or inferred emission factors. You get the comparison you need to defend a self-hosting decision to a CFO and a sustainability officer in the same meeting.
If your AI strategy has to survive both a CFO and a sustainability officer, "the vendor publishes an average" is not an answer. Horizon Platform makes it a trail.
Horizon Platform instruments what you already own; it isn't a carbon-neutral claim, a certification, or an offset programme. It's a trail a sustainability team can use.
Deployment & architecture
Standard container stack: Docker, PostgreSQL, Redis, with secrets managed through environment configuration or an integrated vault. Advanced integrations (identity providers, SIEM, ticketing) and agent-side modules are added based on environment requirements.
Deployment targets: on-premises, private cloud, or a regulated hybrid. Reference architectures available for common environments on request.
Who Horizon Platform is for
It fits when
- Data sensitivity, regulatory scope, or commercial value makes cloud-only AI a non-starter.
- You operate a mixed fleet and want AI-assisted operations on top, not a rip-and-replace.
- Governance and audit matter as much as capability.
It doesn't fit when
- You are committed to a cloud-first, SaaS-only AI strategy and see self-hosting as overhead.
- You don't yet have an AI use case with a justified ROI. Start with the cost calculator or a discovery call first.
See it in your environment
A demo takes about 30 minutes and is scoped to the workflows you care about. We'll walk a real scenario, not a scripted tour.
Next step
Self-hosted AI, with a trail a CFO and an auditor can both follow.
Bring a real workload — operational, security-evidence, or ESG. We'll show you what the platform measures, where the chain of custody lives, and how the numbers come out the other side.
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.
