Skip to content
ArkNet AlphaRemote execution runtime layer

Remote execution
for the workloads that matter now.

ArkNet is a remote execution runtime layer for coding agents, model workloads, and provider-addressable compute. Route work to the right machine across NVIDIA, AMD, and Metal.

Coding agents

Route builds, tests, evals, and task execution onto the machine that fits the work instead of the machine the agent started on.

Model workloads

Run training, evals, and inference across NVIDIA, AMD, and Metal through one addressable runtime path.

Portable execution

Turn heterogeneous machines into one execution surface for teams, labs, and infrastructure operators.

arknet runtime
ARK_PROVIDERS=0x15F4...8B15,0x92A1...44CE arknet run agent_eval.ark
Resolving preferred providers...
[Select] Node-1037 | NVIDIA H100 | ap-northeast
[Metrics] p50 21ms | p95 47ms | reliability 100%
✓ Remote execution context established.
tail -f /logs/exec.log
Running evaluation workload on remote provider...
suite 12/48 — pass rate: 100%
Streaming logs, timing, and results back to local context.

What ArkNet is

ArkNet is the layer between software that wants to execute and the machine that should execute it.

Remote execution runtime

ArkNet separates where software decides from where execution actually happens.

Heterogeneous by design

Target NVIDIA, AMD, and Metal across Linux, Windows, and macOS through one routable surface.

Provider-aware routing

Route by address, region, latency, reliability, VRAM, and pricing with deterministic preference order.

Current use cases

ArkNet is not the use case. It is the execution layer underneath the use cases already emerging.

Agent execution

Use ArkNet as the runtime path for coding agents that need to do real work on real machines.

Run builds and tests remotely
Pin preferred providers with ARK_PROVIDERS
Return logs, latency, and execution outcomes

Training and inference

Dispatch training loops, eval jobs, and inference workloads to the backend and hardware that fit the task.

NVIDIA, AMD, and Metal
Cross-platform runtime path
Addressable provider selection

Provider networks

Expose machines as routable execution capacity with live metrics, reliability, and pricing.

Publish provider addresses
Expose p50 and p95 metrics
Turn hardware into selectable runtime capacity
Provider network

Turn machines into runtime capacity.

Publish a provider, expose metrics and pricing, and become part of a routable execution surface that users can address directly from their projects.

Addressable
Provider addresses
Users can browse the registry, copy provider addresses, and route via ARK_PROVIDERS.
Observable
p50, p95, reliability
Expose the metrics needed to make execution routing legible and trusted.
Portable
NVIDIA, AMD, Metal
One runtime layer across heterogeneous hardware families and operating systems.