Performance for shared workloads
At 6.3 tokens per second per user versus a 0.7 reference baseline, the illustrated throughput is 9×. Across 32 chats, that is 201.6 versus 22.4 tokens per second.
Local & Self-hosted AI for secure chat, agents and knowledge with model governance, DLP and audit evidence for enterprises.
Built for the teams responsible for trust
A workspace for your people. A control plane for your IT team. An evidence layer for security and compliance. All on your infrastructure.
Deploy in your private environment. Choose models and configure the platform around your workflows.
Bring in your identity provider, document libraries and internal systems. Define who can use each model and tool.
Manage policies, review activity, verify network isolation and collect evidence from one administration console.
Interactive software preview · sample data · explore all 30 modules
An operational overview of your AI estate.
Requests
1,284
GPU nodes
3
Ready services
12/12
GPU node health
Based on the Kaldryn Platform admin console. Available features depend on license tier, role, hardware and configuration.
Local hosting is the starting point. Kaldryn adds the identity, governance and operational tools needed to make private AI accountable.
Privacy becomes a workflow.
Limit exposure with local processing, scoped document access and configurable DLP. Review and execute subject-data erasure requests with an audit trail.
Make responsible AI operational.
Make AI use transparent with in-product disclosures, model governance and operational oversight. Keep policies, audit activity and reviewable evidence together.
Bring SAML SSO, role-based access, MFA and passkeys together with network-isolation checks, an egress register and reviewable security evidence.
Kaldryn provides technical controls and compliance-supporting documentation, not automatic regulatory compliance or certification. Your obligations depend on the use case, model, configuration and organisational measures.
Discuss your security requirementsCloud subscriptions and token charges add up across a team. With Kaldryn, local inference has no token or credit bill. Compare your cloud spend with the full cost of running your own AI.
Cost comparison · USD
Include subscriptions and average token or credit spend.
Total Kaldryn cost / month
Fixed at $7,500 per month, with unlimited users to onboard.
Potential annual savings
No metered token or credit charges for local inference.
This comparison uses a fixed Kaldryn budget of $7,500/month. Adjust the team size and cloud spend to reflect your organisation. Cloud defaults are illustrative. Model quality and deployment capacity vary; confirm the configuration and scope in your quote.
Cloud = people × monthly cost × 12. Kaldryn = total monthly cost × 12.
See the potential of shared inference with a conservative 9× comparison for 32 simultaneous chats on GB10 hardware.
the per-user throughput in this illustration
Continuous batching · shared GPU
Reference baseline for this illustration
At 6.3 tokens per second per user versus a 0.7 reference baseline, the illustrated throughput is 9×. Across 32 chats, that is 201.6 versus 22.4 tokens per second.
Chat, document search with RAG, agents and model fine-tuning come together in one platform.
Manage SSO, role-based access, DLP and audit logs alongside GPU health, signed updates and backups in one console.
Illustrative comparison, not a new measured benchmark: 0.7 × 9 = 6.3 tokens/s per user. Reference workload: GB10, Qwen2.5-7B, 8k context, 150-token responses, warm engine, 32 concurrent chats. Actual throughput depends on the model, hardware and configuration.
Short answers for teams evaluating on-premise AI. We run Kaldryn Platform on infrastructure you control.
Kaldryn is a private AI platform from Kaldryn Industries AB in Stockholm, Sweden. It gives an organisation a workspace for chat, agents and company knowledge, and an administration console for models, identity, data policies and evidence. You deploy it on infrastructure your organisation controls.
Local AI, also called on-premise AI, means the models run in an environment your organisation controls instead of in a vendor's cloud. Kaldryn is built for that pattern: on-premises chat, agents and knowledge in the private environment you deploy. The platform is air-gap capable. Whether a given deployment is actually isolated depends on the network, firewall, connectors, backups and support access you configure.
Sovereign AI is local AI with control attached. Kaldryn describes it as sovereign AI with enterprise control: the model, the data and the infrastructure stay with the organisation, rather than being rented from someone else's cloud. The about page states the same idea as a design principle. Data control should follow from where the software runs, not only from a contract clause.
Cloud AI is usually billed as subscriptions, tokens or credits, and it processes prompts in the provider's environment. With Kaldryn, local inference does not require a cloud token or credit purchase. Hardware, electricity, licensing and operations still have costs. Identity, policies, model operations and evidence sit in the same console as the workspace, on infrastructure you control.
Kaldryn includes tooling that supports the work. It does not certify an organisation, and it does not replace a legal assessment. For GDPR, the platform provides local processing, scoped document access, configurable DLP that can block, redact or log, and a reviewed flow for subject-data erasure with an audit trail, plus DPIA and data-governance documentation. For the EU AI Act, it provides in-product AI disclosures, literacy records, a model inventory, trust policies, controlled access and operational logs. NIS2 is one of the European rules covered in Kaldryn's regulation briefings. The same console shows identity, least-privilege access, network-isolation checks, an egress register and audit exports that administrators can use when they document how the deployment is run. The duties still depend on the use case, the configuration and the organisation's own measures.
You choose the infrastructure. Models are installed and operated there, on CPU or GPU nodes, and local model files such as GGUF can be imported. Document retrieval is designed to run in that same environment, and prompts for local inference are processed there. Connectors, backups and remote support can be outbound paths. The console includes isolation checks and an egress register so administrators can review the boundary. Remote support, where it is used, is an explicit, time-limited administrator action with a recorded reason.
Administrators choose, install and operate models on the deployment's own infrastructure. The console includes a model catalogue, a hardware advisor for memory and context requirements, CPU and GPU deployment, and import of local model files including GGUF. Training Studio can adapt a model with LoRA for specialist tasks, and GPU nodes can be enrolled in a cluster. Throughput and concurrency vary with the model, quantisation, context length, hardware and workload. What a deployment can run depends on licence, role, hardware and configuration. Model names shown in previews on this website are examples, not a live customer catalogue.
Enterprises and other organisations that need AI their security, privacy and compliance teams can account for. Kaldryn describes that audience as regulated teams in healthcare, legal, finance, defence and the public sector, across the Nordics and the EU. The about page names hospitals, law firms, banks and public agencies as the kind of organisation for whom sending data to a third-party cloud ended the discussion. The product is built and operated by Kaldryn Industries AB in Stockholm.
Tell us about your company, your users and how you use AI today. We’ll help you compare costs and find the right setup.
Company name and business email required.