Astrid is an open-source operating system designed for AI agents, providing a microkernel architecture with capsules for modular, secure, and scalable AI agent deployment.
Source: Official README and linked documentation View on GitHub →Astrid is being discussed because its innovative approach to AI agent deployment, addressing the limitations of traditional agent frameworks by offering a modular, secure, and scalable architecture. Its unique use of capsules for isolation and composition differs, allowing for offline operation, novel agent architectures, and transparent caching, among other features.
Source: Official README and linked documentationAstrid's capsules are isolated WASM processes that can be swapped out to customize the system without forking the core OS. They are described by a `Capsule.toml` manifest and can declare their imports and exports, enabling a configurable and modular architecture.
Source: Official README and linked documentationAstrid implements a multi-layered security model with hard blocks, capability tokens, budget enforcement, approvals, and audit logging to ensure secure and accountable AI agent operations.
Source: Official README and linked documentationCapsules run in a WebAssembly sandbox with strict limits on memory, time, and access to host resources, providing a secure environment for AI agents.
Source: Official README and linked documentationAstrid's architecture is based on a microkernel design with a fixed kernel and swappable capsules. It uses a virtual filesystem with copy-on-write overlay, an IPC event bus, and a cryptographic audit trail. The architecture emphasizes modularity, security, and isolation, with a focus on scalable AI agent deployment.
Source: Official repository tree and dependency filesastrid-approvalastrid-auditastrid-capabilitiesastrid-buildastrid-capsuleastrid-capsule-installastrid-cliastrid-configastrid-daemonastrid-coreastrid-cryptoastrid-eventsastrid-hooksastrid-kernelastrid-mcpastrid-openclawastrid-preludeastrid-storageastrid-telemetryastrid-testastrid-uplinkastrid-vfsastrid-workspaceAstrid is suitable for AI research and development, enterprise AI deployment, and any scenario requiring secure, scalable, and customizable AI agents. It is useful for creating autonomous agents, novel agent architectures, and for reducing LLM costs through caching.
Source: Official README and linked documentationv0.7.0 (2026-05-25): Moved various structures to `as`, fixed security issues, added `astrid-build` installation, and renamed VFS scheme.
Source: Official releasesAstrid is a project for teams and individuals involved in AI research and development, offering a unique and innovative approach to AI agent deployment. Its modular and secure architecture makes it suitable for a wide range of applications, although it may require a steep learning curve and a solid understanding of Rust and WebAssembly.
Source: Official repository