Show HN: Submilli – runtime with semantic permissions for agents that write code

Category: infrastructure

Tags: agentic-ai, sandbox, permissions, wasm, code-execution, security

Score: 7.8/10 (Innovation: 8, Technical: 8, Documentation: 8, Utility: 7)

Submilli is a Rust-based runtime that executes agent-generated TypeScript code in WebAssembly with a semantic permission model called Blueprints. It enforces fine-grained, per-operation rules (e.g., refund limits per customer tier) outside the model's control, with built-in packages, audit trails, and resource limits. This fills a critical security gap for business agents that generate code.

Target audience: AI engineers, backend developers building agentic workflows, security engineers

Repository: https://github.com/submilli/submilli-runtime · Rust · Apache-2.0 · 28 stars

View on Hacker News