A .muflo artifact is portable across every environment ADAPT supports — the same compiled file runs unchanged whether the host is a browser tab, a Linux server, or a zero-knowledge virtual machine. The evaluator changes shape per target; the artifact and its semantics do not. See Architecture — the C++ evaluator core for why that property is load-bearing rather than incidental.

This page summarises the targets and what each is good for. For the operational details of a specific target, follow the link in the relevant section.

Shipping today

Native (Node.js + N-API)

A server-side evaluator built as a Node.js native add-on. The fastest target; uses libsodium for cryptographic randomness.

Use it for backend services running on Linux x86-64. Server-side ADAPT nodes (storage nodes, aggregator nodes, the message broker) all run on the native target.

Operational page: Native ADAPT wrapper.

Browser / WASM

The evaluator compiled to WebAssembly with Emscripten. Bundled as part of @adapt-toolkit/sdk. No runtime download — the WASM blob is ~4 MB and ships inside the npm tarball.

Use it for front-end nodes embedded in web and mobile UI. Also useful in Node.js when you want byte-for-byte parity with the browser evaluator for testing.

Operational page: Browser ADAPT wrapper.

Demonstrated, on the roadmap

AWS Nitro Enclaves

The native evaluator linked against libnsm, hosted inside an attested Nitro Enclave. The evaluator participates in the enclave’s attestation chain — a remote party can verify it is running the expected .muflo inside an expected enclave configuration without trusting the operator.

Use it for server-side workloads that need hardware-rooted attestation: key management, multi-party computation backends, regulated-data processing.

Operational page: Enclave ADAPT wrapper.

RISC Zero zkVM

The evaluator compiled to RISC-V and executed inside a RISC Zero zero-knowledge virtual machine. Every state transition produces a succinct proof that the transition was computed correctly; anyone can verify the proof without seeing the data the transition operated on.

Use it for scenarios where the result of a computation must be publicly verifiable but the underlying data must stay private — auditing, compliance attestations, public-good computations on private inputs.

This target has been demonstrated end-to-end. First-class tooling and a shipping wrapper are on the roadmap.

Anticipated

Mobile targets (iOS / Android) and additional enclave technologies (Intel SGX, Apple Secure Enclave) are tractable for the same reason RISC Zero was: the evaluator’s surface to its host is narrow enough to port to anywhere the C++ compiler can reach. We add first-class targets in response to customer demand. If you have an environment in mind, get in touch through Contact Us in the top navigation bar.