Cloud-Loader Agent: Seamless Dotfile Migration & AI Knowledge Tracking
What Is Cloud-Loader Agent?
cloud-loader-agent is the backend service powering loader.land — an open-source platform designed for AI developers. It solves the friction of migrating local AI assistant configurations (like CLAUDE.md, AGENTS.md, and .cursorrules) between machines or sharing prompt templates with teams.
Rather than relying on complex OAuth flows or manual Git syncing, Cloud-Loader uses ephemeral 6-character codes for instant file transfer, with built-in auto-deletion after 24 hours. Recently, it has expanded to include a Concept Tracking engine that builds knowledge graphs via AI.
Tech Stack
| Layer | Technology |
|---|---|
| API Framework | FastAPI (uvicorn) |
| Data Validation | Pydantic / SQLModel |
| Python Version | Python 3.12+ (managed by uv) |
| Background Jobs | APScheduler |
| AI / Search | claude-agent-sdk, tavily-python |
| Build System | Hatch (pyproject.toml) |
Architecture & Features
The project is structured as a modern Python backend utilizing pyproject.toml for dependencies and uv for fast package resolution.
Ephemeral Configuration Migration
The core feature allows users to upload zipped backup files of their AI configurations. The system generates a 6-character code which can be used on a target machine to download the payload. Privacy is enforced by an APScheduler background job that purges files and records automatically after 24 hours.
Template Sharing System
Users can share specific configuration files (like a highly tuned CLAUDE.md or AGENTS.md for OpenAI Codex) as templates. These templates have a longer 7-day lifespan, are accessible via JSON or raw markdown endpoints, and track download analytics.
Concept Tracking Engine
The backend integrates tavily-python for automated web search and the claude-agent-sdk to digest that search data into knowledge graphs. This allows users to track specific topics over time and automatically generate content drafts (like tweets or articles) based on the latest context.
What I Learned / Engineering Decisions
A standout engineering choice is the deliberate absence of user accounts for the core migration flow. By relying on ephemeral 6-character codes and aggressive 24-hour TTLs (Time-To-Live) managed by APScheduler, the system reduces friction to zero while inherently minimizing data liability.
For the Concept Tracking features, the architecture transitions to require API key authentication, cleanly separating the public utility endpoints (migration/templates) from the private, compute-heavy AI generation endpoints. The use of FastAPI with SQLModel ensures that the database schema and API payloads remain strictly typed and validated throughout the entire request lifecycle.