Compute
Compute
Eden's Compute category turns laptops, private servers, cloud instances, and GPU workers into a governed runtime surface.
Runtime and Fleet coordinates fleets, nodes, model runtimes, compatibility, placement, shared state, capacity, and health. Gateway provides the governed client boundary for workloads exposed through Eden.
Core concepts
| Concept | Purpose |
|---|---|
| Fleet | A group of runtime nodes operated under common organization and policy boundaries. |
| Node | An enrolled machine or runtime worker with an immutable identity and declared capabilities. |
| Runtime | The software that executes a workload, such as an eligible model server. |
| Compatibility profile | The model, runtime, hardware, protocol, and state-sharing constraints a route must satisfy. |
| Placement | The decision about where work should run based on policy, capacity, latency, privacy, and cost. |
| Desired state | The models, runtimes, capacity, and health conditions the control plane reconciles. |
Capabilities
Workload runtime
Run eligible services and inference workloads across local, private, public, and managed environments without changing the client-facing gateway contract.
Fleets and nodes
Enroll machines with explicit identities, capabilities, compatibility, fleet membership, and allowed gateway principals.
Placement and scheduling
Choose eligible runtimes using policy, capacity, latency, locality, privacy, cost, licensing, hardware, and compatibility constraints.
Model serving
Operate compatible local and vLLM-style runtimes behind the same governed AI gateway used for managed providers. See Run AI.
Capacity sharing
Pool, allocate, or lease eligible capacity without merging tenant credentials, namespaces, policy, or telemetry boundaries.
Runtime health
Track node availability, utilization, drain state, session lifecycle, and failover signals through Operations & Security.
Security boundary
A runtime node is not an unauthenticated public endpoint. Enrollment identity, fleet membership, allowed gateway principals, model compatibility, and lease scope are checked before traffic can use it. Private and marketplace capacity must retain separate credentials, namespaces, and telemetry boundaries.