An operating system for private, open-source AI. Host your models and agents with verifiable privacy, full control over your code and data, and zero lock-in. Our infrastructure framework is open source — so you can run everything on our managed cloud or export and self-host at any time.
Platform Features
Confidential Agent Framework
Run AI agents with verifiable privacy guarantees. Each agent executes inside an isolated confidential VM with its own VPN layer for encrypted transport. Plugins and microservices are fully compartmentalized — a compromise of one cannot reach another.
Private Model Hosting
Deploy open-source models or your own fine-tuned models on encrypted compute. Your model weights are encrypted at rest and in transit using per-tenant keys. No provider — including Covenant — can read your models or your data.
Agent Version Control
All agent templates are versioned, auditable, and rollback-capable. Every deployment is tracked and reproducible. Built for the change management and compliance requirements of enterprise teams.
Open-Source Infrastructure
Our infrastructure framework is fully open source. You are never locked into our managed cloud. Export your agents, models, and entire infrastructure at any time and self-host on your own cluster.
Flexible Model Sources
Agents can use dedicated compute on our network with open-source models and your own models and fine-tunes, or make anonymized calls to centralized providers like OpenAI and Anthropic — all from the same framework.
Verifiable Privacy
Privacy is enforced at the infrastructure layer, not by policy. Our experimental Model Encryption Protocol applies a structure-preserving transform to model weights so inference runs privately on standard GPUs with under 5% overhead.
Model and Compute Options
Dedicated Compute
Single-tenant GPU instances on the Covenant network. Guaranteed isolation and predictable performance.
Open-Source Models
Pre-encrypted library including Llama, Mistral, and Qwen. Ready to deploy immediately, no setup required.
Your Fine-Tuned Models
Upload proprietary models encrypted with your keys. Never visible to other tenants or Covenant operators.
Centralized API Routing
Route to OpenAI, Anthropic, or other providers over encrypted transport. Prompts protected in transit.
Ready to deploy.
Explore the product offering or send us a message to discuss deployment.