Run Untrusted Code Safely Inside Your Application
Execute third-party WebAssembly modules with strict memory boundaries, deterministic CPU step limits, and explicit WASI 0.3 permissions for files and network streams—no root access required.
vvm is a lightweight virtual machine for running third-party plugins, custom business rules, and serverless functions in a strict sandbox. It supports WebAssembly 3.0 and asynchronous WASI 0.3, includes a built-in V-to-WebAssembly compiler (vwasm), starts without JIT warm-up delays, and snapshots or restores memory in less than a microsecond.
.wasm in under 250 ms and runs modules without JIT warm-up delays.
How vvm lets you run customer code and third-party modules safely without heavy containers or 60 MB+ JIT engines.
Execute third-party WebAssembly modules with strict memory boundaries, deterministic CPU step limits, and explicit WASI 0.3 permissions for files and network streams—no root access required.
Clone a pre-warmed sandbox in under a microsecond using 4 KB Copy-on-Write memory pages instead of keeping idle containers running. The entire runtime ships as a ~1.3 MB binary with zero C++ or Rust libraries.
Compile V guest programs straight to .wasm in under 250 ms with built-in vwasm, link modules from Rust, Go, TypeScript, or C dynamically, or build custom rule languages with lexical closures and fair CPU time-slicing.
Runs WebAssembly workloads inside varc, executes transactional stored procedures inside vprevail (rolling back memory automatically if a transaction aborts), filters traffic in vgate, and sandboxes AI tool calls for vllm.
varc Wasm workloads • vprevail procedures • vca module signatures • vgate edge routing.
A look at how vvm's single-pass register runtime and 4 KB Copy-on-Write snapshots compare to traditional JIT engines.
| Runtime Capability | vvm (Wasm 3.0 & WASI 0.3) | Wasmtime / Cranelift | V8 Isolates / Node Wasm |
|---|---|---|---|
| Binary Size & Dependencies | ~1.3 MB binary • Zero C/C++/Rust/LLVM dependencies | 35 MB – 65 MB Rust/C++ binary | 80 MB+ V8 C++ runtime |
| Execution Model | Single-pass 16-byte register instruction engine | Multi-tier Cranelift / Winch JIT | Liftoff baseline + TurboFan JIT |
| Snapshot & Rollback Speed | < 1 µs 4 KB Copy-on-Write pages + WasmGC heap | OS process fork or full 64 KB page copy | Full heap serialization (milliseconds) |
| WebAssembly 3.0 Support | WasmGC, recursive types, try_table exceptions, tail calls, SIMD | Varies by feature flags | Tied to JavaScript garbage collector |
| Cross-Architecture Migration | Move paused instances between x86_64 and ARM64 | Not supported across JIT architectures | Not supported across architectures |
| Debugging & Inspection | Built-in web debugger with step-back replay | External command-line debuggers (LLDB/GDB) | Requires attaching Chrome DevTools |
WebAssembly 3.0 bytecode execution, memory management, and WASI 0.3 runtime specifications.
Validates WebAssembly 3.0 binaries or WAT text and translates them in one linear pass into compact 16-byte register instructions, avoiding operand-stack overhead.
Includes a built-in V-to-WebAssembly compiler (vwasm, < 250 ms), multi-module dynamic linking, WasmGC structs/arrays, and lexical closures in the language builder API.
Implements WebAssembly 3.0 exception handling (tag, throw, try_table, exnref) and constant-stack tail-call instructions.
Supports 128-bit v128 vector math, lane shuffles, integer dot products, and WebAssembly 3.0 Relaxed SIMD fused multiply-add instructions.
Tracks 64-bit Memory64 changes in 4 KB sub-pages for sub-microsecond rollbacks, cross-arch migration, and deterministic multi-tenant CPU time-slicing.
Implements capability-scoped WASI 0.3 async streams and futures, OpenAPI 3.1 endpoints, 18 runnable examples, and a token-authenticated 4-role web debugger.
Follow recent and upcoming runtime milestones, and subscribe for release notifications, early access builds, or product briefings.
vwasm Compiler & Dynamic Linker
Available Now
Single-pass register engine, built-in V-to-WebAssembly compiler (vwasm), multi-module dynamic linker, WasmGC, closures, exceptions, tail calls, and 236 SIMD ops.
WASI 0.3 async streams, sub-microsecond 4 KB memory snapshots, fair multi-tenant CPU time-slicing, x86_64/ARM64 migration, and authenticated 4-workspace web studio.
vgate Edge Filter & Sidecar Integration
Available Now
Embeds directly inside vgate via the vgate.http ABI with 4 KiB CoW per-request isolation and Prometheus metrics, plus OIDC IAP/WAF/SOCKS5h sidecar protection for vvm studio.
Enter your email below to receive release announcements, benchmark notes, and early access builds. No spam.