Covenant Labs provides an open‑source AI operating system that enables private, sovereign AI inference by encrypting model weights and data with per‑tenant keys, allowing encrypted inference on standard GPUs with less than 5 % performance overhead. Their Covenant Cloud platform runs AI agents in isolated confidential VMs and offers provider‑agnostic orchestration, so enterprises can deploy and manage models without vendor lock‑in while maintaining full control over privacy and compliance.
Funding
Funding not disclosed
Founders
Product
Problem
Current AI infrastructure is dominated by centralized providers that retain control over model weights, data, and inference pipelines, exposing users to privacy breaches, lock‑in, and the loss of competitive advantage. Existing privacy solutions such as homomorphic encryption or secure enclaves are either prohibitively slow or limited by hardware constraints, making scalable, private AI deployment impractical for most organizations.
Solution
Covenant Labs offers an open‑source AI operating system that delivers verifiable, cryptographic privacy for models and data while eliminating vendor lock‑in. Their Covenant Cloud platform runs AI agents and models inside isolated confidential VMs with per‑tenant encryption keys, ensuring that neither the provider nor any operator can access the underlying assets. The proprietary Model Encryption Protocol (M.E.P.) encrypts model weights so inference can be performed on standard GPUs with less than 5 % performance overhead, providing near‑native speed without sacrificing security. Conduit, the framework’s infrastructure layer, automates resource allocation, type‑safe schema validation, and multi‑model orchestration across any compute provider, allowing developers to deploy open‑source or fine‑tuned models with minimal DevOps effort. All components are fully open source, enabling customers to run workloads on Covenant’s managed cloud or export the stack for self‑hosting at any time.
Target Audience
Primary customers are enterprises, regulated industries, and developer teams that require secure, private AI inference and want to retain full control over their models and data while avoiding dependence on centralized AI providers.
Features
- Confidential Agent Framework: agents execute in isolated VMs with dedicated VPN layers, compartmentalizing plugins and microservices.
- Private Model Hosting: model weights are encrypted at rest and in transit using per‑tenant keys; providers cannot read the models or data.
- Model Encryption Protocol (M.E.P.): structure‑preserving weight encryption enables encrypted inference on standard GPUs with <5 % overhead.
- Agent Version Control: auditable, rollback‑capable templates ensure reproducible deployments and compliance.
- Conduit AI Infrastructure: type‑safe input/output schemas, automatic GPU/memory sizing, and block‑based pipeline composition.
- Multi‑model, provider‑agnostic orchestration: load‑balanced replica management and runtime abstraction across clouds, on‑prem, or hybrid environments.
- Open‑source framework: full code transparency and the ability to export and self‑host the entire stack without lock‑in.