Kiln

Meld fuses. Loom weaves. Synth transpiles. Kiln fires. Sigil seals.

A Rust implementation of a WebAssembly runtime with full Component Model and WASI 0.2 support. Designed for safety-critical embedded systems with bounded allocations, deterministic execution, and a modular no_std architecture for automotive, medical, and aerospace environments.

Kiln bridges the gap between WebAssembly's portability and the strict requirements of safety-critical deployment. It runs on everything from cloud servers to bare-metal Cortex-M targets.

# Clone and build
git clone https://github.com/pulseengine/kiln
cd kiln
cargo build --bin kilnd --features "std,kiln-execution"
# Run a WebAssembly component
./target/debug/kilnd your_component.wasm --component

• kilnd/ — Runtime daemon (main executable)
• kiln-runtime/ — Execution engine
• kiln-component/ — Component Model support
• kiln-decoder/ — Binary format parsing
• kiln-foundation/ — Core types and bounded collections
• cargo-kiln/ — Build tooling

# Basic component execution
kilnd component.wasm --component
# With WASI support
kilnd component.wasm --component --wasi
# Set resource limits
kilnd component.wasm --component --fuel 100000 --memory 1048576

# Install build tool (optional but recommended)
cargo install --path cargo-kiln
# Build runtime
cargo build --bin kilnd --features "std,kiln-execution"
# Run tests
cargo test --workspace

Early Development — Basic WebAssembly component execution is working:

./target/debug/kilnd hello_rust.wasm --component
# Output: Hello wasm component world from Rust!

• WebAssembly Component Model parsing and instantiation
• WASI 0.2 stdout/stderr output (wasi:cli/stdout, wasi:io/streams)
• Core WebAssembly module execution
• Basic memory management with bounds checking
• no_std compatible foundation

• Additional WASI 0.2 interfaces (filesystem, environment)
• Cross-component function calls
• Full Component Model linking

kiln's correctness rests on conformance testing and bounded model checking — not (yet) a mechanized proof of its WebAssembly execution semantics. We state that plainly.

• WebAssembly spec suite: the current pass rate is re-derived by CI on every run and shown in the badge above (files passed / total · executed-assertion rate), so the number here can't drift from the evidence. We report the file ratio as the headline on purpose: a file that fails to parse (e.g. the custom-descriptors proposal) contributes zero assertions to both sides of the assertion rate, so only the file count reflects it — the assertion rate alone would flatter exactly where it shouldn't. Known gaps: validation strictness on some GC-proposal cases (a few should-be-invalid modules are currently accepted) and the custom-descriptors parse failure. Reproduce locally with cargo-kiln testsuite --run-wast.
• Unit tests: 300+ across the interpreter core.
• Kani (CBMC bounded model checking): proof harnesses on selected safety-relevant components (foundation collections, arithmetic overflow, LEB128, host dispatch, wasi-nn bounds, platform sync) — bounded proofs of those components, not of the interpreter core (kiln-runtime / kiln-instructions / kiln-decoder carry no formal proofs today).
• Not claimed: formal proof of the interpreter's execution semantics. Mechanized semantics (WasmCert-style) is a research direction, not a shipped guarantee.

Other PulseEngine tools use additional techniques (Rocq, Z3/translation-validation, Verus) on their own artifacts; that does not transfer a proof onto kiln's interpreter. These claims are gated by claim-check so they cannot drift from the evidence.

MIT License — see LICENSE.