Sovereign AI You Can Govern: AlchemiStudio and SUSE AI Factory with NVIDIA

Introduction

Artificial intelligence (AI) is spreading across your organization. Business teams use AI assistants to prepare proposals, write marketing briefs, summarize policies and analyze financial data, among many other uses. Developers call models from several providers to build agentic applications. Those agents act across your enterprise systems. Each use creates value, and each raises the same questions. Who is using AI? What are they using it for? Which data does it touch? What does it cost?

AI governance answers these questions. Technology leaders need to enforce organizational AI policy so that AI is used responsibly, in how it is used, what it is used for, which data it touches and what it costs. Governance is not about restricting your users. It sets the boundaries that let them use AI with confidence and streamlines AI use toward organizational goals, so you can approve more of it.

Sovereignty addresses an altogether different question: does your AI stay within your control, with your sensitive data inside your borders? Most organizations today need both. This blog shows how AlchemiStudio, SUSE and NVIDIA come together to deliver sovereign AI that you can also govern.

Customer challenges

Governance challenges

IT needs control of token cost. Every assistant, agent and application consumes tokens, and spend grows with each new use. Unlimited token spend is no longer a plan. IT needs budgets and limits. Without central controls, spend is hard to predict against budget, hard to cap, and hard to attribute to the team or application responsible.

Technology leaders need to enforce organizational AI policy uniformly. CIOs, CTOs and CAIOs decide which models, data and tools each role may use. They need to apply that policy across every team through role-based access control, so AI use supports organizational goals without sacrificing security.

Technology leaders need to define what agents may do before they let them act. Agents do more than answer questions. They work across enterprise systems and can run multi-step and scheduled workflows, including autonomously and while a person is away. To trust agents with real work, IT needs to control which tools and systems each agent can use, set permissions and action policies, require human approval for sensitive actions, and keep a record of what agents do.

Risk, legal and audit teams need evidence. They must show who did what, under which controls, and which records support that conclusion.

Sovereignty challenges

Your AI, your control and your data must stay within your control. Many organizations, especially in regulated industries and the public sector, need proprietary data to stay inside their own borders, and they need the option of privately hosted models and private model endpoints, not only commercial services.

Environments differ. You may run AI in data centers, private clouds, public clouds and you need the same control wherever it runs.

The Solution

SUSE AI Factory with NVIDIA operates your AI: it gives you a secure place to run it, on infrastructure you control. AlchemiStudio governs it: it gives technology leaders the controls to set organizational AI policy, manage cost and show what happened. Together they deliver sovereign AI that you can also govern.

Three parts work together:

• SUSE AI Factory provides the secure Kubernetes, lifecycle, security and observability foundation, managed with SUSE Rancher Prime and protected by SUSE Security, with SUSE Observability for metrics, logs and traces.
• NVIDIA provides the model and graphics processing unit (GPU) layer. AlchemiStudio uses NVIDIA NIM microservices and Nemotron models served from the SUSE AI Library, supported by the NVIDIA GPU Operator and NVIDIA Data Center GPU Manager.
• AlchemiStudio is the enterprise AI operating layer. It gives you one platform to govern, manage and operate AI agents at scale, from business copilots to developer infrastructure, inside your own boundary.

How the solution works

How AlchemiStudio, SUSE and NVIDIA deliver sovereign AI

The reference architecture has two layers. Operate answers “Where does it run, and who holds the keys?” Govern answers “Who decided it should, and can we prove it?”

Your infrastructure. SUSE Linux Enterprise Server or SUSE Linux Micro runs on your own hardware, in your data center, private cloud or edge location. RKE2 or K3s provides Kubernetes, managed by SUSE Rancher Prime, protected by SUSE Security and monitored with SUSE Observability.

Your models. The SUSE AI Library supplies supported AI tools, including NVIDIA NIM and Nemotron. Privately hosted models run on Kubernetes behind private model endpoints, so proprietary data can stay inside your boundary.

Your control layer. AlchemiStudio runs on this foundation, in your cloud, your data center or a disconnected environment. Its model gateway also reaches major commercial model providers, so budgets and access policies cover those requests too, where your policy allows. Your policy sets which models and data classes each team can use, and sensitive workloads can be pinned to in-perimeter or sovereign models.

Your identity, keys, policies and retention settings remain under your control throughout.

The four components of AlchemiStudio

AlchemiStudio has four modules. Your organization builds and approves the agents that run on them.

• AlchemiStudio Cockpit gives IT and security teams control of identity, models, data, agent permissions, budgets, usage limits, approvals and audit evidence.
• AlchemiStudio Console gives engineering teams a model gateway, connectors, observability and workflows, along with virtual keys, prompts, evaluations, traces and logs. Every request through the gateway carries your budgets, rate limits and guardrails, so teams build quickly within the cost limits IT sets.
• AlchemiStudio Codesk lets people in every business function get work done with AI, from drafting a campaign brief or summarizing a policy to analyzing spreadsheet variances or triaging tickets, in the applications they already use. Users can also hand work to agents that run autonomously on a schedule or while they are away. Access follows each person’s role, and sensitive actions pause for human approval.
• AlchemiStudio Compute is a secure runtime for isolated, ephemeral code, file and browser execution.

Control AI costs with budgets, limits and scoped keys

Developers want fast access to the best model for each task. IT wants to know what that costs and to keep it within budget. Cockpit and Console address both.

• Budgets and limits that you set. Set budgets, usage limits and rate limits on scoped keys, with controls for requests, tokens and concurrency. Budgets are enforced before a request runs.
• Spend you can see as it happens. A real-time dashboard shows spend as it accrues, so IT can compare it with budgets and catch overruns early.
• The right model for each agent, application and stage. Choose which models each agent or application may use, and vary them by stage. The same application can use a lower-cost model in development and the model you have approved for production, so experiments stay inexpensive and production stays controlled.
• Approved models without shared credentials. Developers use approved models through the Console model gateway with scoped, revocable virtual keys, so you do not distribute provider credentials.
• Cost attribution. Attribute usage and cost to the teams, applications and agents responsible, so IT can say who owns each cost.

Enforce AI policy across teams, models and tools

Technology leaders need to enforce organizational AI policy uniformly, and AlchemiStudio gives them the power to do it. CIOs, CTOs and Chief AI Officers (CAIOs) set the direction. AlchemiStudio enforces it across every team, model and tool on the platform, based on role-based access control (RBAC) and policy, so AI use stays aligned with organizational goals without sacrificing security. Cockpit turns your policy into controls:

• Identity and access. Control who can use AI through your identity system, with RBAC down to model, tool, sources and connections. Scoped keys can be revoked when needed.
• Models and data. Decide which models each team may use and which data classes they can reach. Policy is applied at request time, not after the fact.
• Agent permissions. Define which tools and systems each agent may use, connect agents to enterprise systems through governed connections, and require human approval for sensitive actions.

Governance of this kind is not about restraining AI use. It sets clear boundaries, so teams move faster within guardrails, and you can say yes to more AI with confidence.

Guardrails, security and audit evidence

Guardrails and approvals. Guardrails inspect inputs and outputs for personal data, secrets, prompt injection, jailbreak attempts, toxicity and restricted topics. Apply them by user, key, endpoint, or agent. Approval steps and permissions govern what agents may do with tools and data, and AlchemiStudio can stop unauthorized requests and actions that pass through the platform’s controls.

Isolated execution. AlchemiStudio Compute runs agent code, file and browser actions in isolated, ephemeral sandboxes that start in under a second. Each run starts from a fresh, identical state with memory separated between sandboxes, and the sandbox is torn down automatically when the task completes. Hard timeouts stop runaway processes.

Runs wherever you deploy. Compute runs any OCI image in the cloud, on premises or fully air-gapped. SUSE Security adds zero-trust workload protection beneath the platform.

Audit evidence. AlchemiStudio records requests, responses, tokens, cost, latency, routing, guardrail outcomes, tool calls and administrative changes, along with approval records. You can export this evidence to your existing observability tooling. It can help address portions of the governance and regulatory obligations facing organizations in financial services, public sector, healthcare, telecommunications and critical infrastructure.

From approved model access to governed business workflows

Here is one example of how the pieces work together:

1. An IT administrator uses Cockpit to set the policy: which models each team may use, the budgets and usage limits, and which agent actions need human approval.
2. A developer receives a scoped virtual key and calls an approved model through the Console gateway, without holding a provider credential.
3. The platform checks each request against the key, model access, budget and limits before it runs, and attributes usage and cost to the responsible team, application or agent.
4. A business user asks a governed agent in Codesk to complete a task. The agent works within its approved tools and data, a sensitive action pauses for human approval, and any code, file or browser work runs in isolation in Compute.
5. The platform records requests, tool calls, guardrail outcomes and approval records, so risk, finance and audit teams can see who did what, under which controls, and what it cost.

An example: preparing a customer proposal

A sales manager asks a governed agent in Codesk to prepare a proposal for an account.

• The agent works only with the information and tools that the manager’s role allows, such as approved company knowledge and connected applications such as Salesforce, Outlook and Slack.
• It uses a model that your policy approves for that team.
• It drafts the proposal. If the next step is sensitive, such as sending it to the customer, the agent pauses for human approval where your policy requires it.
• The manager reviews and approves. The platform records what the agent used and did, and what it cost.

An example: connecting a developer

A developer receives a scoped virtual key that carries approved models and a budget, valid for the period a manager sets, such as the three weeks of a project.

• They point a compatible client or coding assistant at the Console gateway.
• Requests run against approved models and stay within the budget.
• They review usage, cost and traces in Console.

These examples are simplified. AlchemiStudio also handles more complex flows, including multi-step and scheduled workflows with human approval steps, and the same pattern applies across sales, marketing, HR, finance and operations. Talk to us about your scenario.

The table shows how responsibility divides between the people who use AI and the people who govern it.

	Developers and business users	IT and security teams (Cockpit)
Models	Use approved models through scoped virtual keys	Set which models each team may use
Spending	Work within budgets and usage limits	Configure budgets and limits, and track spend in real time
Data and tools	Reach approved knowledge and connections	Decide which data classes and tools each team and agent can reach
Agent actions	Delegate tasks to governed agents and respond to approval requests	Define which tools and systems agents may use, and which actions need approval
Evidence	Inspect traces and logs for the applications they build	Review audit records, guardrail events and exports

What technology leaders, developers and business teams gain

• CIOs, CTOs and CAIOs set organizational AI policy and see how AI is used and what it costs.
• Engineering teams use coding assistants and build agents with approved models, scoped keys and traces.
• Business and operations teams prepare proposals, research accounts, triage tickets and reconcile records with governed agents.
• Risk, legal, compliance and audit teams review AI use, investigate incidents, examine approvals and collect evidence.
• Finance and procurement teams allocate AI spend and see who owns each cost, with real-time spend attribution and budgets.

Next step

Ready to see sovereign AI you can govern? Book a demo. Contact Moses Mathuram, Chief Revenue Officer at AlchemiStudio, at moses@alchemistudio.ai, or Martina Ilieva, Lead, AI Partner Ecosystem at SUSE, at martina.ilieva@suse.com.

Download the SUSE AI Factory with NVIDIA and AlchemiStudio Solution Flyer .

Learn more at alchemistudio.ai  and suse.com . AlchemiStudio is a SUSE partner and a member of NVIDIA Inception.