astrid, from Astrid Runtime, is a portable, capability-secure operating system that connects language models to real-world tool execution. It exposes modular WebAssembly "capsules" via the Model Context Protocol so models can invoke isolated tools while the runtime enforces execution-layer security to prevent unrestricted host access. The runtime offers a CLI, REST API, and MCP server plus SDKs for JavaScript and TypeScript. Targeted at AI engineers and architects, it provides runtime trust and governance for agentic workflows.

What tasks can you actually use it for?

Astrid functions as an Agent Operating System that exposes modular tools to language models, letting models discover and invoke capabilities through MCP. Developers pack functionality into versioned WebAssembly "capsules" that run isolated from the host. Supported capsule languages include:

• Rust
• JavaScript
• TypeScript

How auditable and secure are agent actions?

The runtime enforces a capability-security model and runs capsule code inside WebAssembly sandboxes, placing execution-layer controls between agents and the host. Astrid includes an audit subsystem called Unicity Audit to record tool calls and security-relevant events, and it routes system interactions through a virtualized airlock (VFS). Because enforcement happens at execution rather than in prompts, the tool reduces common prompt-injection attack vectors, although correctness still depends on capsule code and invoked model outputs.

Is it practical to adopt in existing developer workflows?

Adoption fits developer-led environments because Astrid provides a CLI, a self-hosted REST API, or a dedicated MCP server and runs on Linux, macOS, and Windows. The core is written in Rust while SDKs exist for JavaScript and TypeScript, so integrating into web and backend stacks is possible. Teams should plan for a developer workflow that builds, versions, and tests Wasm capsules, since extending agent capabilities requires code artifacts rather than simple prompt recipes.

A practical choice for engineering teams that need execution-layer governance

Astrid suits engineering teams that require controlled agent execution and traceable tool calls, as its design prioritizes runtime trust and auditability. Expect a development-centered integration path: teams must produce and test Wasm capsules and incorporate the runtime's audit signals into governance processes. Adoption requires engineering investment and governance planning.