Pixel-native VM
Deterministic execution model with pixel-addressed representation and image-native program/data concepts.
OPEN SOURCE · DISTRIBUTED COMPUTE
ZDX Parallel / Pyxel VM combines a deterministic pixel-native VM with a private, opt-in distributed compute fabric. Nodes can contribute bounded compute, exchange authenticated work and immutable artifacts, recover from leases and failures, and participate without receiving arbitrary remote host execution authority.
Deterministic execution model with pixel-addressed representation and image-native program/data concepts.
Private, opt-in clusters with karmic/gravitational leadership election, no fixed permanent coordinator, and a 100-node cap per cluster.
CPU, memory, charging/load, process, and guard conditions restrict participation; remote peers are not given arbitrary shell or Python execution.
Ed25519 identity, authenticated enrollment, replay protection, and TLS-oriented transport controls.
Immutable SHA-256-addressed artifacts with AES-GCM encryption at rest, resumable transfer, and task-scoped grants.
Linux/VPS worker tooling plus an Android node implementation with capability reporting, background service support, and result attestation paths.
VERIFIED IN RELEASE WORKSPACE
The repository release verification report records Python compilation and regression checks, authenticated node integration, signed registration, gossip and queue convergence, resumable artifact behavior, lease recovery, resource admission, election/state recovery, and altered-attestation rejection.
EXTERNAL GATES REMAIN
The published verification report still calls for clean multi-host Linux validation, real TLS/mTLS operator deployment, broader physical-network testing, and full Android device/toolchain validation for ordinary VM tasks. Those paths are not represented here as already verified.
ARCHITECTURE PRINCIPLES
Leadership is elected within the cluster rather than hard-wired to one permanent machine.
Workers, artifacts, result attestations, and unsupported Android VM paths are rejected rather than silently treated as trusted.
Worker policies restrict resource use and the system avoids general-purpose remote shell execution.
The VM, distributed runtime, Android code, verification documentation, and release evidence are available in the public repository.