A compute cluster from the organisation's own computers
Purpose
RUNVIK Cluster — an application that unites an organisation's computers into a compute cluster.
Unites an organisation's computers into a single compute cluster: distributes compute jobs and language-model requests across nodes based on speed and link quality, verifies results and isolates faulty or compromised nodes.
Capabilities
Nodes connect outbound with no inbound ports; node keys, signed and numbered messages, TLS
Nodes run only jobs from their registry; built-in and custom job types
Measures node speed and link stability; distributes jobs only when a gain is predicted
Spot re-verification of results, recomputation and node quarantine on tampering; release only after owner biometric confirmation
Single OpenAI-compatible endpoint for node language models with best-node selection and failover
Works through connection drops with result buffering; collaborative model training
Web dashboard of node status, job and event log, API keys; Russian and English
Who it is for
Companies with branches and remote sites, engineering and computing departments, IT, organisations running their own language models on several servers, universities and research institutes.
Delivery and licence
Delivered for installation inside the customer's perimeter: on a server or workstations. Licensed per seat for a selected term; the key is bound to the computer and contains the licensee's tax ID. Includes user guide, installation guide and licence agreement.
Screenshots, video overview and demo access are provided on company request.
Dear users! Ideas on how to improve RUNVIK Cluster? Suggest an improvement or write to support@runvik.ru.