strukto-ai/mirage

Mirage is a virtual filesystem for AI agents, a layer that makes cloud services and data sources look like ordinary folders on a hard drive. Instead of an AI agent needing to know a different API (programming interface) for S3 file storage, Google Drive, Slack, GitHub, Gmail, and each other service, Mirage presents all of them as directories under one unified file tree. The agent can then use familiar Unix-style commands like cp, cat, and grep across all of them as if they were local files. The core idea is that most AI language models already understand shell commands and file paths very well because they were trained on enormous amounts of code and documentation about these concepts. By expressing every external service as a filesystem, Mirage lets AI agents work with a vocabulary they are already fluent in, rather than learning new APIs for each integration. In practice, you define a workspace by mapping paths like /slack, /s3, and /github to the corresponding service connectors. The agent then executes bash commands against that workspace and Mirage handles translating them into the appropriate service calls behind the scenes. Workspaces can be snapshotted, cloned, and versioned, so you can save and restore the full environment state. Mirage ships as Python and TypeScript SDKs and a command-line tool. It integrates with major AI agent frameworks including LangChain, Vercel AI SDK, and OpenAI Agents SDK, and works with coding agents like Claude Code. Python 3.12 or higher is required. The full README is longer than what was provided.