Runtime guides

pd-edge

pd-edge is the edge data-plane runtime and edge ABI for VM programs. It provides an HTTP proxy runtime, an interactive console runtime, and a multi-protocol sample echo server.

cargo run -p pd-edge --bin pd-edge-console -- --help

For HTTP proxy, console, admin API, active control-plane RPC, Docker, protocol DAGs, and release artifacts, use pd-edge and the runtime implementation guide.

pd-controller

pd-controller is the control-plane service for pd-edge data planes. It supplies edge polling and result callbacks, program and debug-session orchestration, health and metrics endpoints, and a Web UI.

cargo run -p pd-controller

The controller listens on 0.0.0.0:9100 by default and serves a built Web UI under /ui. Operational details are in pd-controller.

micro-rustscript

micro-rustscript runs VMBC on ESP32-C3, ESP32-S31 preview hardware, and a native Arduino simulator. ESP32-C3 loads /rustscript/main.vmbc from SD, then its dedicated flash partition, then the serial VMBC REPL.

pio run -e arduino

The Arduino target builds the bridge and a compiled VMBC program on the host. A successful simulator run finishes with rss:status=0. See micro-rustscript for target images, VMBC partition replacement, and the serial REPL.

IronRust

IronRust provides a RustScript runtime and compiler for Microsoft CLR. PdVm.Compiler converts VMBC to CLR assemblies, while PdVm.Runner supports compile, run, and compile-run workflows.

dotnet build IronRust.sln

IronRust requires the .NET 10 SDK. Build the solution first; then use PdVm.Runner to compile or run RSS and VMBC programs. The typed CLR wrapper discovers reachable use System::... imports and generates bindings for supported public members. Dynamic reflection requires explicit --enable-dynamic-dotnet. See IronRust.

Flint

Flint is a RustScript-native AI inference framework. It exposes Torch tensor operations, llama.cpp primitives, stable-diffusion.cpp image generation, GGML discovery, tokenization, and safetensors I/O as host functions.

Use Flint for the current CLI invocation, host namespaces, and model-specific RSS programs. --llm selects Torch-based RSS programs, --llama selects llama.cpp programs, and --sd selects stable-diffusion.cpp or GGML programs. Scripts keep model architecture, sampling loops, and inference workflows in RSS.