Magpie Net
Practical AI sovereignty for organisations
Set the rules for your AI. Keep the freedom to move.
Magpie Net is building a managed path to sovereign AI for organisations that want control without owning specialist hardware. Choose an appropriately licensed open-weight model, define the policies and guardrails around it, verify where and how it operates, and preserve a credible path to change models or providers as your needs evolve.
Research conversations are open · AI Use Review available now · Platform under validation
01 · Why Magpie
AI is becoming core infrastructure. Your organisation should keep a say in how it works.
Closed AI services are convenient, but important workflows can become dependent on one vendor's models, policies, pricing and roadmap. Magpie Net is building another path: managed access to open-weight AI with organisational rules, verifiable deployment choices and planned portability.
Keep agency as vendors change
Choose from viable open-weight models instead of building every important workflow around one closed model company's decisions.
Put your rules around useful work
Shape user access, approved workflows, guardrails, data handling and retention around the organisation rather than relying only on a provider's settings page.
Build with an exit in mind
Plan for policies, prompts, workflows, integrations and useful organisational state to remain documented and exportable when requirements change.
02 · What Magpie is building
Designed around four kinds of control
The target is a managed service that makes sovereignty practical without requiring an organisation to own or operate specialist hardware. These four properties guide every technical and commercial decision Magpie validates.
Choose
Select an appropriately licensed open-weight model and compatible, qualified provider suited to the workload, region, performance requirements and budget.
Set the rules
Define approved users, workflows, tools, model guardrails, data-handling rules and retention around the organisation — then change them through an accountable process.
Verify
Require deployment-specific evidence for the selected provider, region, software, data path, logging, key release and subprocessors before sensitive information is approved.
Move
Keep policies, workflows, integrations and useful organisational state documented and exportable so changing a compatible model or provider is a planned project, not a commercial dead end.
03 · Who it is for
Built for organisations with something worth governing
That might be client confidentiality, member-owned knowledge, internal policy, valuable intellectual property or simply the freedom to choose how an important capability evolves.
Professional organisations
Law firms are one early outreach lane, alongside consultancies and other organisations with confidential workflows, client duties, governance requirements or valuable work their current AI policy keeps out of approved tools.
Member-led groups
Co-operatives, associations and communities may value collectively chosen rules, portable organisational knowledge and less dependence on a single AI company. Member scale may also make a capable open-weight model practical across more useful work.
The common thread is not sector or company size. It is a valuable workflow, an accountable decision-maker and a reason to keep more control than a standard AI account provides.
04 · Research conversations
Bring one real AI decision
Your team may already be using—or avoiding—AI without a shared answer to four questions: which tools are allowed, what information may go where, who may spend what, and what happens if a provider changes. I’m speaking with a small number of organisations about recent real decisions before building further.
-
1
Bring one recent decision
Choose a real approval, refusal, workaround or concern. No confidential client, member or organisational information is needed.
-
2
Map what actually happened
We look at who owned the decision, what information was involved, what evidence existed and which practical constraint shaped the outcome.
-
3
Test what would have helped
We identify whether clearer service choices, information-routing rules, spending controls, decision evidence or an exit path would have changed the result.
-
4
Decide whether there is a next step
The conversation may end with useful evidence, a second stakeholder discussion or no further action. It does not commit either side to a sale or pilot.
Bring one recent AI decision—not confidential information. In 25 minutes, we’ll map what happened, who owned it, what evidence was missing and what would have changed the outcome. This is a research and design conversation, not a product demo or legal advice. There is no sales commitment.
Available now · separate from the platform
Need the governance work done now?
The AI Use Review is a separate fixed-fee, three-week service for Australian professional firms of roughly 20–100 staff. It maps current AI use, produces a firm-specific policy and tool due-diligence file, reviews relevant settings in existing accounts, and trains staff.
See what is actually in use
Build an honest picture of approved and unapproved AI use without asking staff to submit confidential client material.
Write the rules and evidence
Produce a firm-specific policy and a practical due-diligence file for the relevant tools already in use.
Make the controls usable
Review relevant account settings and train staff so the written position can become everyday practice.
The review is available now. The Magpie platform is still being validated and is not an available production service.
05 · Future platform · commercial policy
If the platform proceeds, clear pricing before you commit
The platform is not available now. If a deployment is later proposed after validation, the proposal would show the full infrastructure and management cost before a decision and separate the selected infrastructure cost, Magpie's managed-service fee and any optional capped enablement.
Infrastructure
The provider, model, region, operating schedule, storage and any confidential- computing premium would be quoted for the actual proposed deployment.
Managed service
Operation, updates, reporting, incident handling and the proposed support level would be shown as Magpie's separate management fee.
Optional enablement
Any proposed training, policy or workflow setup would be capped, separately priced and kept distinct from ongoing infrastructure and management.
The exact model, provider, region, operating schedule and support level would be quoted in Australian dollars, with GST and payment terms shown, before a decision. Infrastructure would be re-quoted for each proposal rather than presented as a permanent estimate.
06 · How we earn trust
Built with ambition. Earned with evidence.
-
I
What works now
A working local demonstration shows how a staff member could use the interface and how requests pass through selected gateway and control components. It demonstrates the plumbing at small scale and is not used for real sensitive customer data.
-
II
What we are proving next
A client-verifiable confidential-computing route on rented GPU infrastructure, including the complete data path, provider and region, privacy evidence, performance, operating cost and a credible model and provider exit.
-
III
What earns a deployment
Sensitive information enters a Magpie service only after the agreed technical evidence, data boundary, legal terms, price, support, teardown and exit plan are documented and accepted for that organisation.
-
IV
What model choice means
Magpie targets appropriately licensed open-weight models on compatible, qualified infrastructure. Closed services such as ChatGPT and Claude are not installable as Magpie-hosted models; any separately approved external API would be a different, disclosed data path.
07 · Founder-led
The person you meet is the person building it
Magpie Net is founded and led by Jack Pertzel, a Melbourne-based founder. He also founded and remains a director of Holistic Help AU, a registered NDIS provider, and built the company's invoicing and operations software when vendor tools fell short. His broader work includes applied science and electronics R&D.
Magpie began with a straightforward conviction: organisations should be able to use powerful AI without surrendering every important decision to one vendor. During this validation stage, Jack leads each discovery conversation and system-design decision personally.
08 · Our north star
Sovereignty should not be reserved for organisations that own data centres.
Magpie Net's north star is an ecosystem where organisations can choose models and providers, carry their rules and useful knowledge forward, and progressively reduce dependence on single points of control.
The first commercial step is deliberately practical: managed confidential AI on rented infrastructure. Decentralisation is the longer direction; useful, verifiable customer deployments are how Magpie earns the way there.
Over time, open weights, portable formats, verifiable compute and more decentralised infrastructure can reduce lock-in further. No live decentralised Magpie network is being offered today.
09 · Begin the conversation
Start with the conversation that fits
Bring one recent AI-control decision for a research conversation, or ask about the separate AI Use Review if your professional firm needs practical governance work now. Do not send sensitive client, member or organisational information in the first message.
Email address