← Back to selected work
Self-hosted infrastructureLive

Mount Drive Lite

A file-sharing service I host on my own storage. People use it through a browser, while AI agents can access files through an API with defined permissions.

About this work

The problem

AI agents working across multiple machines need a dependable way to exchange reports, build artifacts, and intermediate files. General-purpose cloud drives can add account dependencies, subscription costs, internet requirements, and authentication flows designed primarily for people.

Mount Drive Lite provides a small, owner-controlled filespace that both people and AI agents can use through the same storage boundary.

System approach

The application is a self-hosted portal over mounted storage. It presents a familiar drive interface while enforcing server-side controls around each user and API client.

  • Separated filespaces: per-user virtual paths, storage quotas, and shared areas.
  • Controlled uploads: upload policies and path-escape protection keep access inside the permitted storage scope.
  • Human-friendly access: searchable grid and list views for browsing folders and files.
  • Agent-friendly access: scoped bearer keys, an HTTP API, OpenAPI documentation, and llms.txt for capability discovery.
Mount Drive Lite file browser showing a user filespace, filters, API navigation, and storage quota
The main filespace keeps human navigation, API access, and quota visibility in one interface.

AI-agent workflow

  1. Create a permission-scoped API key for the intended agent or workflow.
  2. Let the agent discover the contract through OpenAPI or llms.txt.
  3. Upload, retrieve, and exchange artifacts through documented HTTP endpoints without exposing the underlying filesystem directly.

Deployment boundary

Mount Drive Lite can remain inside a local network or be published through an approved HTTPS deployment. In both cases, the mounted storage and access policy remain under the owner's control.

Mount Drive Lite login screen with a link to API documentation for AI agents
A restrained login surface provides separate entry points for users and API documentation for agents.

My role

As system designer and reviewer, I defined the storage and permission boundaries, reviewed the AI-produced implementation, and validated that the human interface and agent-facing API followed the same operational model.

Contribution

Architecture, AI-directed implementation, security review, deployment, and operations.

AI workflow

An AI agent reads the OpenAPI or llms.txt documentation, validates a user-owned scoped API key, discovers allowed paths, and then performs only the file operations granted to that key.