Astrid is a portable, capability-secure operating system for composable software built on WebAssembly and Rust.
Astrid addresses the fundamental security problem of running untrusted agent code on your machine: traditional agent frameworks rely on the model to follow instructions, but Astrid enforces security at the runtime level instead. It treats each component as a sealed WebAssembly capsule with explicit, cryptographically signed capabilities. The kernel is deliberately minimal—it routes events, enforces capabilities, runs the sandbox, and maintains an audit trail, but holds no model logic, tool schemas, or business logic. A compromised capsule cannot read files, access the network, or spawn processes outside its granted capabilities because the kernel enforces these boundaries, not the agent itself.
Astrid suits teams building agent systems where security isolation is non-negotiable: multi-tenant deployments, systems handling sensitive credentials or files, or environments where you need to run untrusted third-party tools. It is independent of any particular LLM provider, agent framework, or user interface, so you choose your own distro and compose it explicitly. The tool's approach differs fundamentally from frameworks that embed security in prompts or trust models to self-limit; instead, it provides OS-grade isolation through a capability model where every file path, network host, and tool is a signed, scoped grant that the kernel verifies before allowing access.
Development on the project shows consistent activity across multiple dimensions. The codebase receives regular updates and maintenance across its core components. The project maintains active documentation including a dedicated architecture book and contributor handbook. Pull requests and issues are addressed with attention to the security-critical nature of the runtime. The team demonstrates engagement with the capability-security model through refinements to the kernel and capsule lifecycle management.
