Run models inside the boundary where your mission already lives.
Aegis is a Rust-native inference server for private, air-gapped, and sovereignty-sensitive environments that cannot delegate trust to a black box.
Serve models close to protected data with provenance, tamper-evident audit, and operator-grade controls.
Run inference where the data already lives.
Aegis is not a cloud wrapper. The page should read like a hardened local runtime: model integrity before boot, OpenAI-compatible serving after boot, and evidence written even when the environment is disconnected.
~$ vortex aegis serve --model ./models/llama-guard --profile enclave[ok] Resolving aegis contract...[ok] Bound signal: Trust boundaryPOST /v1/chat/completionsAuthorization: Bearer $LOCAL_AEGIS_KEY{ "model": "local/llama-guard", "messages": [{ "role": "user", "content": "Classify this record." }], "provenance": { "require_attestation": true }}Artifacts: model integrity manifest · local request audit · attestation envelope- Load approved models into a protected inference runtime.
- Expose an OpenAI-compatible local API to existing clients.
- Write provenance and request evidence without depending on cloud access.
- model integrity manifest
- local request audit
- attestation envelope
Serve models close to protected data with provenance, tamper-evident audit, and operator-grade controls.
Deployment briefing ↗Primitives your teams ship against.
Concrete behaviors and interfaces so security, platform, and product teams share one operational picture.
Private inference with a stable engine abstraction
OpenAI-compatible serving for existing developer workflows
Model provenance and integrity checks before runtime use
Local audit trails designed for disconnected environments