Skip to content
Olympusx AIOlympusx AIOlympusx AI documentation

What Olympusx AI is

Olympusx AI is an AI platform that you install on Kubernetes clusters you own. It serves open-weight models on your GPUs or CPUs, grounds agents and assistants on your own knowledge, and gives every team one gateway to call them through, with keys, token budgets, guardrails and an audit trail.

It is built for organisations that must keep data, models and records inside their own network: banks, insurers, public bodies, healthcare, and any team whose regulator, customer or own policy asks where the data went.

Two halves

Section titled “Two halves”

Olympusx AI has two halves, and the console shows each person the half their role opens.

Half Who uses it What it is for
Platform Platform administrators: the team that installed Olympusx AI Registering clusters, putting GPU and CPU nodes into service, bringing model weights in, choosing the serving runtimes, creating tenants and giving them capacity, prices and billing, platform settings.
Workspace The people in each tenant Deploying granted models, building knowledge, agents, tools, skills and assistants, running and approving their work, judging answers, and governing people, keys, budgets, guardrails, usage and the audit trail.

A tenant is one isolated customer of the platform: a business unit, a subsidiary, a client. Its models, knowledge, agents, keys, conversations and records are kept apart from every other tenant’s. One account belongs to exactly one tenant. See Tenants, projects and people.

What a tenant can do with it

Section titled “What a tenant can do with it”
  • Call models through one gateway. Any application or tool that speaks the OpenAI API calls Olympusx AI by changing its base URL and key. See The gateway and the API overview.
  • Deploy models on its own capacity. Open-weight LLMs, embedding models and rerankers, on GPU or CPU, inside the capacity the platform granted. See Models and deployments.
  • Ground answers on its own documents. Collections in vector databases that belong to the tenant, filled by pipelines, with answers that cite their sources. See Knowledge and grounding.
  • Build agents. Agents that reason, call tools and skills, hand work to collaborator agents, and stop for a person’s approval before the calls you mark. See Agents, tools and skills and Human approval.
  • Publish assistants. An agent published to the tenant’s people, or behind a branded link for people with no account. See Assistants.
  • Turn on use cases. The Code Assistant (editors connected to the tenant’s models, code search over its repositories, merge request review) and Security (assessments of code, investigations of logs and the audit trail). See Code Assistant and Security.
  • See and govern what happened. Runs, traces, conversations, scores, usage in tokens, and an audit log whose entries are hash-chained. See Observability and evaluation.

What stays on your network

Section titled “What stays on your network”

Olympusx AI runs entirely on your clusters. It does not call home, and it needs no route to the internet to run. Prompts, answers, documents, embeddings, traces and the audit log are stored in the platform’s own database and object store, on your infrastructure.

Data leaves your network only when your platform administrator connects something outside it: a cloud model provider whose models a tenant then calls, a code host, a log store or a mail relay. Each of those is a choice made in the console, and each is visible there. See Sovereignty and data.

Where to go next

Section titled “Where to go next”