VoisX Orchestrate
VoisX Orchestrate is an AI orchestration platform that you install and run on your own infrastructure. It runs in your Kubernetes cluster, serves models you choose, and puts one policy checkpoint in front of every model call. It can run on a network with no internet access, with limits on building new model images.
These docs cover how to install and run Orchestrate, how to build with it, and its API.
Who it is for
- Teams that cannot send data to a cloud AI service. Prompts, documents and transcripts stay in your cluster. Calls to external AI providers are off.
- Sites with no internet access. The platform images come in the release bundle and run from your own registry, the licence is a signed file checked locally, and upgrades arrive as signed offline bundles.
- Platform teams. One Model Gateway for every model your organization runs, with policy checks, PII guardrails and an audit log in front of it.
What you can do
| You want to | Read |
|---|---|
| Understand the parts and how a request flows | Architecture and concepts |
| Make your first model call on an installed platform | Quick start |
| Find your way around the admin console | Use the admin console |
| Give people access, with your own identity provider | Users, roles and projects and Identity and single sign-on |
| Work in Arabic, French or other languages | Languages |
| Add, build and deploy your own models, size them and share GPUs | Models and serving, Sizes and GPUs and Your own models and containers |
| Call models from your own code | Call the Model Gateway |
| Build an agent with tools and a knowledge base | Agents |
| Build a multi-step workflow with branches, parallel lanes and loops, and start it on a schedule or a webhook | Workflow graphs and Schedules and webhooks |
| Bring an agent written with ADK, LangGraph, LangChain, CrewAI, the OpenAI Agents SDK or LlamaIndex | Code agents |
| Let an agent read and write your systems, with a person approving writes | Tools and connectors, MCP and OpenAPI and Human approval |
| Talk to an agent by voice, or by phone, in a voice you choose or clone | Voice agents, Voices and speech models and Phone calls (SIP) |
| Put an agent in front of applications, websites, other agents, MCP clients and Teams, Slack or WhatsApp | How agents are exposed and Remote agents and trusted installs |
| Control what is allowed, and what it costs | Policy, PII guardrails, Guardrail profiles, Model access and Usage controls |
| Prove what happened | Audit log and Compliance |
| Test agents before and after they go live | Evals, Live versions and the release gate and Online evals and feedback |
| Call the platform from scripts | API reference |
Install and run it
Orchestrate installs with a Helm chart into a Kubernetes cluster you run.
- Check your cluster against the requirements.
- Install Orchestrate.
- Add your licence.
- Configure exposure, sign-in, models, voice and the features you want.
- Keep it running with the operations guide, including high availability and backups, and upgrade with signed offline bundles.
Deployment modes
| Mode | What it means |
|---|---|
| Air-gapped | No internet access. Platform images come from your own registry. External providers stay off. Building new model images is limited, because the in-cluster build downloads base images and packages from public sources. See Requirements. |
| Self-hosted, connected | Your cluster, on a network that can reach the internet, so model builds can download base images and weights, and channels can reach Teams, Slack and WhatsApp. Calls to external AI providers stay off unless you turn them on. See External providers. |
| Hybrid | A control plane, in your cloud or ours, with site agents on customer premises that run the work touching sensitive data. See Hybrid sites. |
| Multi-organization (optional) | One install hosting several organizations, isolated from each other. See Multi-tenancy. |
For licensing, sizing and pilots, contact us.
Was this page helpful?