
• Secure computation
• Data in Use Protection
• Runtime Engine
Secret-shared computation means inputs stay private between parties, but the answers don't. Write in a Python-like language, compile to bytecode, let the VM handle the protocols. No plaintext to breach, subpoena, or explain to a regulator
What is Stoffel VM?
A register-based virtual machine built specifically for multi-party computation. It executes programs where sensitive data stays secret-shared, and only agreed results leave—no plaintext exposure, no central honeypot.
Inputs stay private
Secret sharing splits data into useless-alone pieces. The VM computes on shares, never plaintext.
Answers only
Raw data and intermediate steps never exist in cleartext. No "we deleted it after" promises.
Production-grade
Register-based architecture with deterministic execution, p50/p95/p99 observability, and protocol-aware scheduling. Designed for systems where performance characteristics need to be auditable, not estimated
Built for real workloads

Separate clear/secret registers
Public data in clear registers, secret-shared values in secret registers. The architecture enforces the separation.

Deterministic execution
Same inputs produce same outputs every time. Reproducible runs satisfies compliance auditing requirements and makes debugging tractable.

Deep observability
Full execution traces, latency percentiles, and instruction-level hooks. Actually debuggable MPC.
How it actually works
Write in StoffelLang
Python-style syntax with explicit secret and public types. You focus on business logic, not protocol implementation.
Compile to bytecode
The compiler generates VM-compatible bytecode with separate handling for clear and secret values. Nothing sensitive can accidentally become readable without an explicit reveal.
Execute with MPC
