You set the rules. You own the stack. Every user, every agent, every model — governed by policies you define and no one beneath you can override.
Hover each tier to see how policies connect across your deployment.
Your entire Helix deployment. Master policies, data residency, approved models, compliance posture. Nothing overrides this level.
A tenant (a client, a business division, an institution). Inherits Universe rules, adds its own. E.g. Finance Galaxy enforces HIPAA on top of company-wide SOC 2.
A business unit, team, or department within a Galaxy. Scoped tool access, model limits, budget caps, persona libraries.
A project, workflow, or application. Inherits everything above. Users here cannot change the rules — they can only work within them.
Policies flow DOWN. Users cannot override what their Galaxy, Solar System, or Universe has set. Companies control their reality.
Explore Universe Architecture →Every governance decision is made by a Council — a circle of humans and AI advisors who vote, log, and enforce policy at every tier of your deployment.
Every request evaluated against Open Policy Agent rules before execution — deny policies enforced in milliseconds with zero model invocation.
Every council vote, policy decision, and blocked request written to a tamper-evident log. Exportable to SIEM — Splunk, Elastic, Datadog.
Councils exist at Universe, Galaxy, and Solar System tiers. Lower councils can only restrict — they can never override the tier above them.
Before any data crosses a boundary — cloud model, remote tool, or external API — Helix scans, detects, and redacts. The scrubbed version travels. The original stays home.
40+ PII entity types: names, SSNs, emails, phone numbers, IBANs, medical record numbers, IP addresses, and custom regex patterns.
Define organisation-specific patterns per Galaxy or Solar System. Healthcare Galaxies flag MRNs; Finance Galaxies flag account numbers.
PII replaced with reversible tokens before external processing. Tokens swapped back locally after response — cloud providers never see real data.
Even on-premise tools in a different Solar System cannot receive PII from another unless an explicit cross-system policy allows it.
Helix runs your private models on your hardware, and connects to cloud providers when policy allows — always through the PII scrubber, always logged, always council-approved.
Task complexity score, cost policy, latency budget, and model capability flags determine which model runs. The router never guesses; it enforces.
The Synapse MCP Router is Helix's unified tool orchestration layer. Models discover tools dynamically — every call intercepted, policy-checked, and logged before the tool fires.
Models discover all available tools via a single manifest — no hard-coded integrations, no restart required.
Every call policy-checked before the tool fires. Deny rules enforced in milliseconds with full audit logging.
Tools register and deregister live. New capabilities instantly available to all models in the Solar System.
REST, gRPC, WebSocket, and event streams — all routed through a single governed gateway.
Visual, event-driven automation. Triggers, conditions, loops, HTTP calls, database queries — all policy-governed.
Autonomous agent crews for long-running research, analysis, and action pipelines. Agent roles scoped by RBAC.
Webhooks, Kafka-compatible event topics, and real-time trigger chains across your entire Universe.
Any workflow step can require human approval before proceeding. Gates are enforced by policy, not by convention.
K3s for appliance deployments; full K8s for enterprise clusters. Namespaced workloads per Galaxy. Helm chart deployment. ArgoCD/Flux GitOps. Auto-scaling inference pods.
H100 / A100 / L40S support. MIG partitioning for multi-tenant GPU sharing. DCGM health monitoring. CUDA-optimised inference serving. vGPU for virtualised environments.
Nutanix AHV · VMware vSAN · Dell VxRail. Helix runs natively on your existing HCI investment. Zero new hardware vendor required. Storage policies enforced per Galaxy namespace.
Every Helix deployment ships with enterprise-grade platform capabilities as first-class citizens — not add-ons.
SAML 2.0, OIDC, LDAP/AD, MFA, FIDO2, JIT provisioning. SSO at Universe level cascades to all Galaxies. Per-tenant identity isolation.
Hierarchical roles, ABAC attributes, OPA enforcement. Roles defined at Universe level are immutable in child scopes. Resource-scoped permissions down to individual tool calls.
Canary rollouts, A/B model routing, emergency kill switches per model or tool. Flags scoped per Galaxy, Solar System, or Planet. Instant propagation.
Prometheus metrics, Grafana dashboards, Alertmanager. Per-Galaxy resource quotas. SIEM export. GPU DCGM integration. Fleet health without crossing tenant boundaries.
ArgoCD / Flux for declarative infrastructure. HashiCorp Vault for secrets, model credentials, API keys. Model version pinning with one-command rollback.
Internal service catalog, OpenAPI specs, software templates, TechDocs. Onboard new Galaxies with pre-approved templates in minutes.
Corrections, approvals, high-quality outputs flagged automatically and queued for training.
Efficient fine-tuning on your GPU nodes. Nightly runs, no manual intervention. Base models stay pristine; adapters layer your domain knowledge.
Fine-tuned adapters progress through staging → canary → production gates. Rollback in one command.
Whisper STT (multilingual transcription) + Piper TTS (natural speech synthesis). Voice-activated agents, meeting transcription, accessibility-first interfaces.
ComfyUI for image generation and analysis. Document OCR. Chart and diagram interpretation. All processed on your GPU fleet.
Ingest PDFs, Word docs, spreadsheets, wikis. Per-tenant RAG pipelines with chunking, embedding, and retrieval scoped to each Solar System.
Every Helix deployment is yours from the ground up. Your models. Your policies. Your rules.