Are you an LLM? You can read better optimized documentation at /docs/on-prem/onprem.md for this page in Markdown format
Jozu Hub Enterprise Edition
Jozu provides DevSecOps for AI/ML models with secure packaging, policy control, and deployment integrity. Built on the open-source CNCF KitOps standard, Jozu brings enterprise-grade governance to self-hosted AI deployments.
NOTE
Contact the Jozu team for an on-premises Proof of Value (POV) engagement. Email [email protected] for more information and scheduling.
Why Enterprises Choose Jozu On-Premises
Fully Private or Air-Gapped
- Installs completely behind your firewall
- Uses your existing registries, RBAC, and authentication systems
- No data leaves your environment
- Works in air-gapped environments
Built on Open Standards
- No vendor lock-in - uses OCI-compliant container registries
- Uses KitOps ModelKits (a CNCF project with 120K+ downloads)
- Leverages the industry standard CNCF ModelPack specification (authored by Jozu)
- Production usage across global enterprises and government agencies
7x Faster Model Deployments
- In-cluster deployment caching eliminates redundant container builds
- Optimizes GPU utilization and reduces infrastructure costs
- Tested with Llama 3.2 8B: 44.9 seconds vs standard deployment of 342.3 seconds
Enterprise Security and Compliance
- Automated security scanning for every model (multiple evaluations)
- Policy enforcement that blocks non-compliant deployments
- Tamper-proof packaging with SHA-based attestation
- Use enterprise-hardened and approved container images for deployments
- Complete audit trails for regulatory reporting (EU AI Act, ISO 42001, NIST AI RMF)
Enterprise Benefits
Automated Risk Reduction
- Prevent unauthorized deployments - SHA-based digest verification blocks tampered or unintentionally changed models
- Security scanning - Automated vulnerability assessment for all model artifacts
- Reproducible deployments - Immutable model packages from development to production
- Audit readiness - 87% reduction in compliance prep time with downloadable audit reports
Increased Velocity & Efficiency
- Faster iterations - 41% improvement in AI delivery velocity
- Reduced manual work - Auto-generated Kubernetes manifests and container images
- Simplified rollbacks - Version-controlled model packages enable reliable rollback procedures
- Team productivity - Data scientists keep existing tools but with added DevOps controls
Cost Optimization
- Higher GPU utilization - 7x faster deployments reduce idle compute time
- Infrastructure efficiency - Cached deployments minimize startup time for models and agents
- Operational overhead - Eliminate custom deployment scripts and manual security checks
Key Features of DevSecOps for AI
Jozu users want to leverage their proven DevOps tools with their new AI projects. The Jozu platform is built to bring the AI/ML and software engineering worlds together.
Immutable & Signed Packaging
Jozu's ModelKits use a Kitfile recipe to package models, datasets, code, and configuration together in an immutable, signed artifact that is stored in an enterprise container registry. ModelKits can be deployed locally or to any serving platform.
This brings the repeatability and traceability of docker to the AI/ML ecosystem.
yaml
# Kitfile example
manifestVersion: v1.0.0
package:
name: fraud-detection-model
version: 2.1.0
model:
- name: fraud_model.pkl
path: ./models/
datasets:
- name: training_data.csv
path: ./data/
code:
- name: inference.py
path: ./src/
Security Scanning and Attestation
Jozu scans every ModelKit for vulnerabilities, but customers can add additional evaluations and scans.
Scan results are available in Jozu, but also attached to the ModelKit as signed attestations - customers use SHA-digests, signing status, and attestations to control access and deployments with existing Open Policy Agent (OPA) tools.
Deployment Automation
Jozu turns any ModelKit into a deployable container or Kubernetes manifest. Customers can add their own preferred / approved base containers, then wire Jozu into their pipelines to guarantee hands-free and secure deployments to development, staging, and production.
An in-cluster cache means deployments from Jozu to Kubernetes are up to 7x faster than normal - reducing the idle time for expensive GPUs and speeding iterative model updates.
Audit and Governance
A complete lineage is kept for each model or dataset packaged by Jozu. View the change history or download an audit report at any time.
Everything is stored in a customer's container registry so it always aligns to their authentication and authorization definitions.
Works With Your Existing Infrastructure:
Jozu can integrate with any tool that works with containers. For example:
- Container registries: GitLab, Artifactory, Nexus, Harbor, DockerHub
- Authentication: Keycloak, Okta, Active Directory
- Orchestration: Kubernetes, OpenShift
- CI/CD: Jenkins, GitLab, GitHub Actions
Built on KitOps Open Source
Jozu is built on KitOps, a CNCF project that provides the foundational CLI and Python SDK for ModelKit packaging:
- 120,000+ downloads since launch in February 2024
- CNCF project - governed by the same foundation as Kubernetes
- Production deployments across global enterprises, government agencies, and research institutions
- Active community with contributors from Red Hat, Digital Ocean, Xebia, Rapid Data, and other major organizations
The ModelKit format used by KitOps directly inspired the ModelPack open specification, making it the first and only reference implementation of this emerging standard.
Proof of Value Engagement
Jozu on-premises is not available for self-installation. We provide a guided 3-week Proof of Value (POV) to demonstrate impact in your environment:
Week 1: Installation and Baseline
- 1-hour installation with your DevOps team during a remote session
- Baseline measurement of current deployment times, security gaps, and audit readiness
- No disruption to existing AI/ML workflows
Week 2: Evaluation
- Real-world testing with your models, teams, and infrastructure
- Performance measurement comparing Jozu to current processes
- Integration validation with your registries, auth systems, and CI/CD pipelines
Week 3: Results Review
- Quantified impact assessment showing deployment speed improvements and risk reduction
- ROI analysis based on your specific deployment patterns
- Implementation roadmap discussion
Infrastructure Requirements
Minimum requirements:
- Kubernetes v1.3.x+ (AWS t3.xlarge or equivalent)
- ReadWriteOnce StorageClass for ephemeral model unpacking
- Postgres database (AWS m5d.large or equivalent) - or use built-in instance
- OCI-compliant registry (Artifactory, Nexus) - or use Jozu's built-in registry
- OIDC-compatible authentication (Keycloak, Okta)
- DevOps/SRE contact with Kubernetes and registry configuration access
Getting Started
To schedule your Proof of Value or discuss Jozu's fit for your organization:
Email: [email protected]
Include information about:
- Your current AI/ML deployment challenges
- Number of models deployed monthly
- Compliance or security requirements
- Preferred timeline for evaluation