Zecheng Zhang
924ec49796
A unified virtual filesystem for AI agents. Mount S3, Google Drive, Slack, Gmail, GitHub, Linear, Notion, Postgres, MongoDB, SSH, and more behind one filesystem so agents read, write, and pipe across services with familiar shell commands. Ships Python (mirage-ai) and TypeScript (@struktoai/mirage-*) SDKs, a CLI, FUSE mounts, and adapters for OpenAI Agents SDK, Vercel AI SDK, LangChain deepagents, Pydantic AI, CAMEL, OpenHands, Mastra, Pi Coding Agent, plus FUSE-based integration with Claude Code and Codex. Apache 2.0 licensed.
52 lines
2.1 KiB
Plaintext
52 lines
2.1 KiB
Plaintext
---
|
|
title: VFS View
|
|
description: How Mirage represents external systems as one unified filesystem. The eyes for AI agents.
|
|
icon: eye
|
|
---
|
|
|
|
## One View Across Systems
|
|
|
|
Mirage gives agents one way to see external systems.
|
|
|
|
Instead of exposing every backend through its own API shape, Mirage represents services, data, and infrastructure through a unified filesystem interface. The agent gets one environment to browse, inspect, and reason about, with no mental-model switch per resource.
|
|
|
|
This is the **eyes** for AI agents: how they perceive what exists, where it lives, and how to reach it.
|
|
|
|
## Why Representation Matters
|
|
|
|
Large language models work best when the environment is legible.
|
|
|
|
Mirage makes systems legible by representing them through:
|
|
|
|
- mounted paths
|
|
- directory structure
|
|
- stable, file-shaped resources
|
|
- familiar discovery commands such as `ls`, `find`, `grep`, and `cat`
|
|
|
|
This is not a UI choice. It is the representation layer that lets agents understand what exists, where it lives, and how the pieces relate.
|
|
|
|
## A Unified Interface
|
|
|
|
Mirage doesn't ask the agent to remember one shape for GitHub, another for S3, another for Slack, and another for local disk.
|
|
|
|
It maps all of them into the same visible environment. Discovery becomes consistent, provenance becomes traceable, and cross-system reasoning gets dramatically simpler.
|
|
|
|
For an agent, `/github/repo/README.md` and `/s3/reports/q1.csv` are different resources, but they live behind the same interface, and the same `cat`, `grep`, or `ls` works on both.
|
|
|
|
## Why This Fits Agents
|
|
|
|
Agents need more than raw access. They need an environment they can inspect and navigate reliably.
|
|
|
|
A unified view gives them:
|
|
|
|
- one way to discover what is available
|
|
- stable paths they can revisit and cite
|
|
- clearer provenance for outputs
|
|
- less prompt overhead spent translating between resource-specific APIs
|
|
|
|
## Mental Shortcut
|
|
|
|
Mirage turns external systems into something agents can read the way Unix users read a filesystem.
|
|
|
|
That unified view is how agents see. It is the eyes through which the [Hands](/home/design/hands) act, the [Arm](/home/design/arm) reaches, and [Recall](/home/design/recall) remembers.
|