Any hardware. Any location. Subnets define the boundary. The network handles everything underneath.
Architecture
The network treats every machine as capacity. Different architectures, different vendors, different locations — unified into a single compute layer.
GPUs, CPUs, ARM, x86, accelerators. The network discovers capabilities automatically. You don't choose instance types — you describe the work.
Cloud, on-prem, home, colo. Hardware joins from anywhere. The network resolves geography so workloads land on the best available machines.
Deploy once, run anywhere. The same interface regardless of what's underneath. No vendor-specific SDKs. No region selection. Just work in, results out.
Anyone can build. Anyone can provide. No approval process to get started. The network is permissionless by default, controlled where it matters.
Boundaries
Open participation or dedicated boundaries — on the same platform. You choose the level of control.
Open compute anyone can use.
Start on the free tier, upgrade when you're ready. The fastest way to start building and attracting providers.
Dedicated compute you control.
Run on hardware that only serves your subnet. Stronger isolation for sensitive workloads. Full control over who provides and what runs.
Define who participates and how.
The organizing layer on top. Every subnet sets its own rules for providers, developers, and workloads. Shared for reach. Isolated for control.
Products
Keep your product surface where it already lives. Move the expensive compute to idyl. Let demand build a subnet around it.
Your frontend, API, database, auth — they stay where they are. Nothing changes about how users interact with your product.
Inference, media processing, batch jobs, training — the heavy work runs on the network. Same workload, fraction of the cost.
As usage increases, providers join your subnet. More providers means more capacity and lower costs. The network scales with your product.
Scale
Idyl is designed to grow with you. A subnet with ten machines works the same as one with a million.
Imagine a community of a million people, each contributing a machine to a single subnet. That's a million computers working together on one mission — research, inference, simulation.
Scale is now practical.
As networks grow, coordination overhead increases. Performance becomes harder to maintain.
Subnets grow with the communities behind them.
Design
The platform handles connectivity, scheduling, and discovery. Subnets define purpose, rules, and access — shaping how compute is used.
Nodes come and go — that's expected. The network adapts automatically, shifting workloads to available hardware so nothing stops running.
The control plane matches workloads to the best available resources in real time. No manual placement, no region selection — the network places work where it runs best.
Performance stays consistent whether a subnet has ten machines or a million. The architecture scales linearly, so growing a subnet never means slowing it down.
Home machines, lab hardware, devices behind NAT — they all connect without extra setup. No STUN, no TURN, no port forwarding. The network reaches everywhere.
Deploy workloads or shape a network. The platform handles the rest.