Now accepting pilot partners
Energy AI that advises. People decide.
An AI-native control platform for the renewable grid, guided by ozari+: intelligence that learns every shift and never operates. Plus the engineering to put it to work.
Illustrative simulation of a sunny day in the NEM. Not live data.
0 AI control paths
2 people approve each command
19.8M fuzz runs, 0 crashes
5.7 s failover in the lab
0 third-party Go packages
The grid changed
The grid changed. Most control rooms did not
Power now flows both ways. Demand can fall below zero at midday. Batteries set prices, and attackers target the devices where plants meet the grid. Operators need clearer data, faster decisions and stronger defence, without handing control to a black box.
51%
Renewables’ share of energy in Australia’s main grid (NEM) in Q4 2025, the first quarter above half.
−263 MW
South Australia’s record-low operational demand, on 25 December 2025. Rooftop solar produced more than homes and businesses were using.
30+
Wind and solar farms hit in one day in Poland in December 2025. Some grid-connection devices were reached with default passwords.
8 months
Average time to sign a NEM connection agreement, up from 2 months in FY22. Engineering capacity is now a constraint.
Platform
One AI-native platform for control, intelligence, defence and data
Ozari runs as one hardened system on your own servers. Every capability shares the same safety gate, the same identity and the same audit trail.
Pilot-ready
Ozari Grid
The AI-native control room for networks full of solar: real-time control, outage restoration and DER orchestration.
In AI-EMS
ozari+
Self-improving grid intelligence that forecasts and explains. The breaker stays open to it: it never operates.
Early access
Cyber defence
Security built into every deployment, plus Ozari Defend, where people authorise every OT response.
Available
Data Exchange
One open data layer for field protocols, CIM/CGMES models and historians, behind a fault-isolated gateway.
Services
Engineering
Controls, SCADA, protocol, OT security and commissioning engineering, from the team that builds Ozari.
ozari+
Meet ozari+. Your people, plus
ozari+ is the AI inside Ozari. It watches the network, forecasts load and solar, finds anomalies and explains them. When it suggests an action, a person raises the command and a second person approves it.
- No command path. Its output has no command field and it runs in an isolated enclave.
- Every answer is explained. Each suggestion carries a confidence score and a plain-language reason.
- People and interlocks decide. Two-person approval, safety tags, a single-writer control lease and a tamper-evident audit apply to every action.
This follows the December 2025 joint guidance from Australia’s ACSC, the US CISA and partner agencies on AI in operational technology, which advises limiting active AI control of OT without a human in the loop.
A day on a feeder
One recommendation, from forecast to outcome
Scroll through one sunny day on feeder F3. Watch ozari+ see a problem coming, two people decide, and the network stay inside its limit.
-
Sunrise. Power flows out
The substation feeds about 1,800 homes. Load is picking up for the morning.
-
Rooftop solar rises
Solar on the roofs now covers most of the street's demand. The flow on the feeder falls toward zero.
-
ozari+ sees it coming
Reverse flow will pass the feeder's 3.0 MW limit at 12:40, and peak at 3.2 MW at 13:10.
ozari+ · confidence 0.91 -
The network model works out the options
Curtail rooftop solar on F3 by 6% from 12:30 to 14:00. Or do nothing, and pass the limit for about 70 minutes.
DERMS engine -
One person asks. A second approves
Operator A raises the command from the advice. Operator B approves the exact command at 12:26. Safety tags are clear and the control lease is held.
Two people · every command -
Held under the limit
Export peaks at 2.9 MW. No one loses supply, and every home keeps most of its solar.
-
Back to normal
The curtailment is lifted through the same two-person path.
-
Evening peak, and a lesson
Power flows out again. ozari+ scores its forecast against the metered flow and recalibrates for tomorrow.
ozari+ · learns from every outcome
Illustrative scenario on the Ozari lab network. Simulated values.
Self-improvement
Self-improving. Never self-approving
ozari+ gets better with every shift, inside strict limits. It can learn and it can propose. Only people can put a new generation into service.
- Observe Score every forecast and log every decision and outcome. Available now
- Adapt Recalibrate to your network’s own history. Available now
- Propose Draft improvements to its own models and rules. In development
- Prove Replay history and run in shadow beside the current generation. In development
- Promote People approve. Versioned, recorded and reversible. In development
The operator console
One operating picture for the whole network
A fast, calm console built for long shifts. Click through the real screens: live energisation, self-healing, forecasts and a verifiable record of every action.
-
01One operating picture
Load, frequency, alarms and customers served at a glance. On the right, ozari+ sums up the network, shows how sure it is, and lists what it advised before and what happened next.
-
02The network, live
The single-line shows energisation as it happens. Solar and wind feeds are green, and normally-open ties are marked. Click to inspect. Right-click to operate, through two-person approval.
-
03Self-healing, advisory by default
When a feeder faults, self-healing (FLISR) finds and isolates it, then proposes restoration through a tie. It stays advisory until an administrator arms it.
-
04Forecasts you can check
ozari+ forecasts load and generation for the next 24 hours and plots what really happened beside it. Every forecast is scored, so errors are visible.
-
05Every action on the record
Commands, approvals and alarm actions land on a SHA-256 hash-chained ledger. The console checks the chain and shows the result, so tampering is visible.
-
06Nothing hidden
A live view of the control plane: every engine, interface and data flow, with its health. ozari+ runs beside the engines, with no route to the command path.
Screens from the Ozari console, showing simulated data from the Ozari lab network.
Autonomy
Autonomy, earned in stages
Self-healing, voltage control and DER curtailment move from off, to advise, to confirm, to automatic only when you arm them, zone by zone, with rate limits and a kill switch. ozari+ stays advisory at every stage.
Stage 0
Off
The default. Ozari watches and records, and does nothing on its own.
Stage 1
Advise
Ozari proposes the action and the reason. An operator raises the command.
Stage 2
Confirm
Ozari prepares the command. An operator confirms it before anything moves.
Stage 3
Auto
Pre-authorised by an administrator for a zone. Interlocks, the lease and the audit still apply.
Cyber defence
Built to run isolated, on your own terms
Ozari is one small binary, with zero third-party dependencies in its control plane. It runs on your servers, inside your zones, with nothing that has to call home. Ozari Defend adds AI-native defence across the wider OT estate.
- On-premise or air-gapped, with an offline, signed licence.
- ozari+ in its own enclave, designed to sit behind a one-way data diode.
- Fail-closed. With a broken audit chain, a full disk or stale data, it stops and says so. It never guesses.
- Designed for IEC 62443 zones and conduits, with an evidence map for your auditors.
Data exchange
Integrate. Don’t rip and replace
Ozari connects to the devices and systems you already run, through one open data layer. Certified protocol stacks work behind a fault-isolated gateway, so a failing stack cannot stop the control plane.
- IEC 61850
- IEC 60870-5-104
- DNP3
- Modbus TCP
- MQTT Sparkplug B
- OPC UA
- CIM / CGMES
- ICCP / TASE.2
- OCSF
- MITRE ATT&CK for ICS
- ISA-18.2 alarm metrics
- Aligned to IEC 62443
Engineering
Engineering for the renewable grid
The NEM will need over 57,000 skilled workers to build and maintain its energy infrastructure by 2050, up from about 36,000 in 2025. We add engineering capacity where projects stall: controls, protocols, security and commissioning.
Source: AEMO 2026 Integrated System Plan
Controls and SCADA
System architecture, point databases, HMI and alarm design, historians and redundancy.
Substation automation
IEC 61850, IEC 60870-5-104, DNP3 and Modbus integration, gateway design and protocol testing.
OT cybersecurity
IEC 62443 zones and conduits, asset inventory, secure remote access and incident readiness.
Testing and commissioning
Factory and site acceptance tests, shadow runs and supervised cutover, with evidence you can hand to an auditor.
Evidence
Measured, not asserted
We publish how we test Ozari and what the tests found. Hostile inputs, crashes, failovers and unauthenticated requests are all part of the routine.
19.8M
Fuzzing executions against the protocol parsers. Zero crashes.
5.7 s
For the standby server to take control after a simulated crash of the primary.
0
Third-party code dependencies in the control plane, and zero data races under the race detector.
401
The answer to every read, including the live stream, without a valid session.
Internal verification results from the Ozari development environment, June 2026. They are not a third-party audit or certification.
Where Ozari fits
Built for the parts of the grid that changed most
Platform + engineering
Distribution networks
Rooftop solar, home batteries, reverse flow and record-low demand. Ozari Grid gives operators the view and the controls.
Engineering now · module in development
Solar and wind generators
Controls, data and security engineering for plants today. The Ozari generation module is being built with design partners.
Engineering now · roadmap
Storage and hybrid plants
Batteries are 53% of the capacity waiting to connect to the NEM. We help design, integrate and secure them.
Early access
Critical infrastructure defence
Sovereign, AI-native cyber defence for OT, with a person approving every response.
How we start
Prove the value before anything moves
Every engagement starts read-only. Ozari runs beside your current systems, sees the same data and issues no commands. Control comes later, in supervised stages, and only if you choose it.
Insights
Plain-English analysis
-
AI that advises, people who decide: what the new AI-in-OT principles mean for control rooms
Cyber agencies now advise limiting AI control of OT without a human in the loop. What that means for control rooms, and…
-
Open standards over point-to-point integration: IEC 61850, CIM, OPC UA and CSIP-AUS in plain English
Custom links lock you in and split your data. A plain guide to the open standards that let mixed-vendor fleets share data,…
-
Poland, December 2025: lessons for every wind and solar site
Attackers hit more than 30 wind and solar farms in Poland through grid-connection devices, some with default credentials. What every site can…
Start safely
Begin with a read-only shadow pilot
Ozari runs beside your current systems, sees the same data and issues no commands. You measure the value with your own data before anything changes.