Frontier.
For the 1% that can't be WASM.
Frontier is the advanced capability profile of Omega — Firecracker-based MicroVMs for workloads that fundamentally cannot live inside a WASM sandbox. Browsers running V8 JIT, desktops with X11/Wayland, GPU inference with CUDA. It's the kernel boundary you reach for when the kernel boundary is the requirement.
Use Frontier when
you must.
The default Omega capability is Standard — WASM via Wasmtime. ~99% of workloads we see are well-served by Standard, including most container apps converted via OCI→WASIX. Reach for Frontier when:
- · You need a JIT compiler that can't run inside WASM (V8, JVM, .NET).
- · You need a kernel boundary specifically (untrusted multi-tenant code at large scale).
- · You need GPU passthrough or a graphical desktop with full GL/Vulkan.
- · You need to host an unmodified Linux VM image (regulatory cargo, legacy systems).
Hard isolation
without the cost.
Firecracker 1.7 with snapshot resume; cold boot is 380–600ms.
Statically linked, seccomp-filtered, single CPU thread per VM.
x86_64 and aarch64 guests. KVM-backed on both.
VFIO + IOMMU. NVIDIA, AMD, and Apple Silicon Metal (where supported).
Choose Frontier
on a per-brane basis.
[brane.browser] substrate = "gaia" capability = "frontier" replicas = 64 [brane.browser.guest] image = "noble-aarch64@sha256:7a4f..." memory_mib = 2048 cpu_quota = 1.5 gpu = "passthrough" [brane.browser.snapshot] resume_from = "snap:browser:warm-cdp:b3:9af2" [brane.browser.network] egress = "outbound-only" allowlist = ["*.wikipedia.org", "api.openalex.org"]
Frontier is a capability flag, not a substrate. Gaia and Biome both accept Frontier branes; with Frontier is rejected at plan time.
selects the base image. Snapshots are signed and content-addressed; the boot path is just a content hash + manifest binding.