amla-sandbox is a WebAssembly-based sandbox for executing LLM-generated code with capability-based security. Unlike popular agent frameworks that use subprocess or exec() (exposing systems to prompt injection attacks), it provides memory isolation through WASM and enforces explicit tool access constraints. The sandbox supports

5m read time From github.com
Post cover image
Table of contents
Why this mattersSecurity modelQuick startJavaScript API NotesLangGraphArchitecturePrecompilationConstraint DSLTradeoffsLicense

Sort: