VoisX
VoisX Orchestrate · self-hosted AI platform

Your own AI platform. On your hardware. Even with no internet.

Serve your own models, build agents with real tools, and govern every call, all inside your Kubernetes cluster. Nothing calls home: the licence is checked locally and upgrades install offline.

Talk to salesSee the architecture
  • Air-gapped ready
  • OpenAI request format
  • Kubernetes
orchestrate / governance / auditChain verified
Sample governance audit log entries
TimeEventResult
09:41:07chat · qwen2.5 · agent/claims-triageallowed
09:41:09pii.scan · credit_card (Luhn)observed
09:41:12tool · rest.update_claim (PATCH)needs approval
09:42:30approval · j.ali@ministryapproved
09:44:02release · claims-triage v7 · evals 48/50blocked
Each entry is hashed with the one before it; edits break the chain.

Government & defence

Networks with no outbound internet, where every update has to be carried in and checked.

Banks, insurers, health

Prompts and customer data stay on your infrastructure; external AI providers are off by default.

Platform teams

One API for every model your organisation runs, in the OpenAI request format, with policy in front of it.

Capabilities

Everything an AI programme needs, run by you.

Models

Serve your own models

Deploy vLLM, Triton, Ollama, Whisper and Piper on Kubernetes from a catalog. Import weights instead of downloading them.

Gateway

One API for everything

Chat, embeddings, speech-to-text and text-to-speech in the OpenAI request format, with exact and semantic caching of chat answers.

Agents

Agents, runs and traces

Versioned agents, a playground, and a trace of every step each run took.

Knowledge

Knowledge bases

Ingest your documents into a private vector store your agents can search.

Connectors

Tools with a human check

SQL, REST, GraphQL, Salesforce and Office 365. Tools can require a person to approve before they run.

  • Salesforce logo
  • Office 365 logo
  • PostgreSQL logo
Voice

Self-hosted voice agents

Real-time speech-to-text, model and speech in one loop, entirely on your servers.

Governance

Policy on every call

Policy rules check every model call, PII guardrails scan chat and speech calls, and blocked requests are logged.

Audit

Tamper-evident audit log

A hash-chained record of who did what, with one-click verification.

Evals

No release without passing evals

Test suites graded by a judge model you choose. A project can block any version that fails.

Architecture

Every request passes the same checkpoint.

CALLERS
Console · Agents · Your apps
SIGN-IN
Built-in sign-in (Keycloak)
MODEL GATEWAY
Policy
PII check
Cache
Writes to a hash-chained audit log
SELF-HOSTED MODELS
vLLM · Triton · Ollama
SPEECH
Whisper · Piper
OFF BY DEFAULT
External providers

Deployment

Install it once. Upgrade it offline.

Licences are signed files checked locally. Upgrades ship as signed bundles that voisctl verifies, refusing downgrades and checking every image after install.

Air-gapped

Available

Platform images come from your own registry; licence and upgrades work offline. Building new model images needs internet.

Self-hosted, connected

Available

Your cluster on a network with internet, so model builds can download images and weights. External providers are off by default.

Hybrid

Planned

Managed control plane, your data plane. On the roadmap.

Coming next

Visual workflow builder · Hybrid deployment · Usage metering

Ask about the roadmap

Orchestrate licensing

Licensed per deployment, sized to your install.

We scope a pilot on your hardware, then license the platform for production.

Talk to salesHow licensing works