Private SI cloud
Dedicated superintelligence (SI) environments for enterprises, deployed and operated by Rillor, with your models and data on infrastructure you control. Run self-hosted LLMs and agents on-premises, in your colocation space or on dedicated GPUs, with open-weight models or commercial model families.
Three ways to deploy
The same environment runs on your servers, in your colocation space, or on dedicated GPU capacity we source and contract for you.
What runs on it
The workloads we deploy and operate inside your environment.
LLM serving for open-weight and commercial model families, with model versions pinned and every change logged.
Search and answers over your own documents through retrieval-augmented generation (RAG). Each citation is checked against its source passage before it reaches the user.
Agents that plan, call tools and hand work to each other, with each step logged for review.
SI assistants reach business tools through MCP servers with per-tool permissions and audit logs. Our Glesso service runs governed tool access in production.
GlessoAgents we build for your own work, run in the same environment.
Agentic systemsModels and agents are tested on your tasks before release, with the same protocol we use in our SI evaluation work.
SI evaluationAgents are tested against direct and indirect prompt injection and excess privilege, with the attacks we run in our agentic security work.
Agentic securityKubernetes and GitOps deployment
Every environment is declared in version control and deployed from it, so each change is reviewed and can be rolled back.
Distributed storage
Block, object and file storage for models, documents and search indexes, with snapshots for recovery.
Distributed storageNetwork-policy isolation
Each tenant and workload gets only the network paths it needs. All other traffic is denied by default.
Tenant isolation testsMonitoring and upgrades
We watch health, capacity and model latency, and apply platform and driver upgrades on a schedule agreed with you.
Our own products run on a platform we built and operate this way.
Security and control
Self-hosted LLM security: you decide who reaches the environment and what leaves it.
- Data locationYour data stays in your environment. Models come to the data.
- IsolationEach tenant runs in its own isolated space, with its own network rules and credentials.
- Access controlRole-based access for people, and scoped credentials for agents and tools.
- Audit logsSign-ins, changes and agent tool calls are logged with time, identity and result.
- TermsData handling, retention and access terms are set per engagement.
We test agents against attack and limit what they can reach before they go live.
Agentic securityThe hardware to run it
If you need hardware, storage or capacity, we source it and arrange how you pay for it.
- GPU systemsNew and secondary GPU systems, specified, tested and delivered, with networking and rack integration.
- Distributed storageBlock, object and file storage for datasets, checkpoints and model weights, next to your GPU systems.
- GPU financingLeases, loans, sale-leasebacks and contract-backed financing, arranged with lending and leasing partners.
- Data-center capacityPower and space for your GPU systems, contracted ahead of need.
We size the GPUs from five inputs: the model's size, the precision it runs at, the context length, how many users it serves at once and your latency target. Model size and precision set the memory the weights take. Context length sets the memory each request holds while it runs (its KV cache), and the number of users served at once multiplies it. The latency target sets how many requests each GPU serves at once.
Size your deploymentHow a deployment runs
One team from the first workload review to daily operation.
Scope
We map the workloads, data, users and constraints, and choose a deployment model.
Design
We size the hardware, platform, models and access rules, and agree the terms.
Deploy
We build the environment, connect your data, and test models and agents before go-live.
Operate
We run, monitor and upgrade the environment, with change records you can review.
Common questions
Does it run on Kubernetes?
Yes. Every environment runs on Kubernetes. It is declared in version control and deployed from it, so each change is reviewed and can be rolled back.
Can it run on servers we already own?
Yes. It runs on your GPU servers in your own facility or in your colocation space, and Rillor deploys and operates it there.
Which models can it serve?
Open-weight and commercial model families, with model versions pinned and every change logged. Models are tested on your tasks before release.
How do you size the GPUs?
From the model's size, the precision it runs at, the context length, how many users it serves at once and your latency target. If you need the hardware, we source it and arrange how you pay for it.
Research
The newest articles from our Insights library, each with a full research report and its sources.
- Security risks in self-hosted LLM deploymentsSelf-hosted LLM security in six documented failure modes: exposed inference servers, serving software flaws, model files that run code, shared KV caches, poisoned documents and GPU container escapes.Private SI cloud7 min readPDF report
- Private LLM deployment: who controls whatOn-premises LLM deployment, colocation, dedicated GPU capacity and managed APIs each place the hardware, facility, data, model weights, logs and operations with different parties.Private SI cloud7 min readPDF report
- How much GPU memory LLM inference needsLLM inference memory is weights plus a KV cache that grows with context and users, with the arithmetic for a 70B model on 80, 94 and 141 GB GPUs.Private SI cloud7 min readPDF report
Notices
SI outputs. Superintelligence (SI) outputs can be wrong. Important decisions should include human review.