Sovereign AI, engineered to be certifiable.
Orvelis builds the governed AI front door for government, critical infrastructure and defence: one place for staff to use AI, deployed inside your own cloud account, where every model, tool and action runs under rules you set and every step is recorded.
- Model
- Harness
- Your data
- Your infrastructure
The model is the easy part.
Anyone can call a model. Making AI dependable on sensitive work, inside the rules your organisation answers to, is an engineering discipline. That discipline is what we do.
Our engineers come from command and control and autonomy software for space and defence, where the line between what a system proposes and what a person authorises is the whole design problem.
A harness is the layer above the model: what it can see, which tools it may use, which actions need a person’s approval, and the record of what it did. The model proposes; the harness decides.
- Harness engineering
- Permission gates, sandboxed agents and approval steps that set what a model may see, do and say. Policy is enforced in code, not left to a prompt.
- In-tenancy deployment
- Single-tenant, inside your own cloud account in an Australian region, with the model gateway, the data and every record the platform creates held there. Open-weight models run locally when data cannot leave the building.
- Assurance and audit
- Guardrails, content filtering and grounding checks you can configure and test yourself, re-verified whenever a model changes, feeding one authoritative, exportable audit trail.
- Decision support
- AI that helps people work and decide, never decides for them. Agents are deny-by-default and approval-gated, and every action is attributed to the person who authorised it.
Principles
Private
Your data, and every conversation, prompt, response and audit record the platform generates, stays on systems you designate. Nothing leaves unless you decide it should.
Sovereign
Australian-owned intellectual property, designed, built and supported onshore. Your account, your region, your keys, and able to run without a single call offshore.
Ownable
Data and configuration export in open formats without our help, and models are substituted by configuration. You own the capability rather than renting a platform you cannot leave.
Certifiable
Architected to the Information Security Manual controls an assessment tests, with documentation and test evidence built alongside the software, on a defined pathway to IRAP PROTECTED.
One governed front door to AI.
Many models behind it, one set of rules in front. Staff get a single workspace for knowledge, drafting, analysis and bounded agent tasks. The organisation gets visibility and control of every AI interaction.
- Available now
- In development
- On the roadmap
Available now
Grounded knowledge
Retrieval over segregated knowledge bases with citations back to the source, and memory that carries context across sessions.
Available now
Any approved model
Claude, OpenAI and open-weight models, including through Amazon Bedrock, added or substituted by configuration. Prompts, routing and guardrails stay portable.
Available now
Agents with a person in the loop
Semi-autonomous agents that are sandboxed, deny-by-default and approval-gated, with reusable skills your teams write once and share.
Available now
Apps from a sentence
Dashboards, forms and small web applications generated from natural language, previewed live and refined in place.
Available now
Testable assurance
Prompt-injection and jailbreak controls at the harness layer that complement the model gateway’s own, grounding checks, and a complete record of every prompt, response and tool call.
Available now
Browser and cloud
The same workspace on the desktop or in a browser, with its backend running as a cloud service. No client installation for browser users.
In development
Enterprise integration
Single sign-on through your identity provider, Microsoft 365 and SharePoint first, then ServiceNow and SAP, with every tool call passing the permission gates.
In development
Administration and cost
A console for users, roles, the model catalogue and policy, with usage and cost reported by user, team and model.
On the roadmap
Shared workspaces
Conversations, knowledge bases, skills and generated apps held as team assets. Asynchronous sharing first, then real-time co-editing with attribution.
On the roadmap
Workflow automation
Successful ad-hoc work packaged into reusable flows, scheduled or triggered by events, built visually by people who do not write code.
On the roadmap
Spatial decision support
Common GIS formats overlaid on an interactive map, with datums handled correctly and every output labelled for analysis and decision support only.
On the roadmap
IRAP PROTECTED
PROTECTED-aligned controls in place before production, and independent assessment completed before the platform holds data at that classification.
Readiness as at September 2026. We mark every capability plainly: what works today, what is being built, and what comes next.
Built with you, one gate at a time.
Not a fixed-scope build that grows by variation. We stand the harness up inside your environment and add capability in small, committed increments, each proven in a sandbox and promoted only when it passes gates for adoption, assurance and cost agreed with you in advance.
-
Gate 0
Sandbox
The harness in an isolated environment in your account, connected to your identity provider and first data sources, with an assurance baseline.
-
Gate 1
Pilot
A small group of real users on use cases defined up front: what the AI may reach, what it may do and what needs a person’s approval.
-
Gate 2
Cohort
Wider rollout once the gates are met, with capacity tested at the next step and the assurance suite re-run before promotion.
-
Gate 3
Scale
More connectors, teams and capability, each added the same way. Every gate is a real decision point: continue, adjust or stop.
Because everything exports in open formats, the cost of leaving approaches zero. That is the measure of value we would rather be judged on.
Jaunt is our harness, made into a product.
An agentic workspace for research, documents, knowledge and the everyday work around them. It runs on the desktop, or with its backend in your cloud and its interface in a browser, and it is the core we build enterprise platforms from.
Your choice of model
Open-weight models on your own hardware, or cloud models on your terms, swapped by configuration.
A cited knowledge vault
One searchable library in open formats that people and agents both read and write, with answers cited to their sources.
Skills and plugins
Teach it your workflows, or extend it with your own tools.
Every step recorded
Each document read, tool action and approval written to an open-format record you keep.
Tell us what has to stay private.
We will show you on working software how it runs, where it can live inside your environment and what it would take to certify.
Start with an account on Jaunt and we will be in touch.