Awesome AI Agents › AI Agent Operating System

astrid-runtime/astrid

⭐ 10272 Rust added to this list on 2026-07-13 repository created 2026-02-15

Astrid is a novel operating system designed for running composable software, specifically emphasizing secure execution for AI agents. It treats each software component as a "capsule" similar to how an operating system handles a process. These capsules are sealed WebAssembly units, capable of being composed, granted explicit authority, and replaced without expanding their inherent reach. The core principle of Astrid is to establish an OS-grade security boundary, which contrasts with traditional agent frameworks that rely on prompt-based trust. The system's security model is built on several layers: a cryptographic capability model with signed, revocable grants; a WebAssembly sandbox that isolates capsules from ambient authority; a 'dumb' kernel that merely routes events, enforces capabilities, and runs the sandbox without holding any business logic; per-principal isolation for data and access; and a signed, hash-linked audit chain for tamper detection. Communication between capsules occurs via an event bus, with dependencies resolved through manifests. The host ABI leverages the WebAssembly component model with versioned WIT packages for tightly controlled interactions. This distributed security approach ensures that authorization is enforced independently and where effects occur, making it highly robust against various attack vectors. Astrid components include a CLI, the `astrid-daemon` kernel, `astrid-build` for capsule compilation, and `astrid-emit` for bus bridging. It supports a live capsule lifecycle, allowing for dynamic updates without service interruption.

https://github.com/astrid-runtime/astrid

agent-runtimeai-agentscapability-securityllmmcpmicrokernelrustsandboxwasmwebassemblyoperating-systemcomposable-softwaresecuritydistributed-securityaudit-chaincomponent-model