Agent-Ready Architecture

AeroFTP is built for both humans and AI agents. As agentic AI becomes the standard way to interact with computers, file management across servers and cloud providers is a fundamental capability that every agent needs. Yet most file transfer tools were designed exclusively for human interaction through GUIs.

AeroFTP bridges this gap with two native access modes: a graphical interface for humans, and a structured CLI for autonomous agents.

The Agentic Era

AI agents like Claude Code, Open Interpreter, Cline, Aider, Devin, Codex CLI, Cursor Agent, Windsurf, Goose, SWE-agent, AutoGPT, and CrewAI operate by executing shell commands and reading their output. Computer use agents (Anthropic, OpenAI Operator) go further by controlling the entire desktop. MCP servers (Model Context Protocol) expose structured tools to LLMs.

All of these need to move files between local machines, remote servers, and cloud storage. AeroFTP gives them a single command (aeroftp-cli) that works across the AeroFTP provider set.

How Agents Use AeroFTP

Any tool that can execute shell commands can use the AeroFTP CLI. The key features that make it agent-friendly:

Vault-based credentials: the --profile flag resolves credentials from the encrypted vault at runtime. Agents never see passwords, tokens, or keys. They only know the profile name.

# Agent connects using a profile name. Password stays in the vault.
aeroftp-cli ls --profile "Production" /var/www/ --json

# Fuzzy matching works
aeroftp-cli ls --profile "prod" /var/www/ --json

# List all available profiles
aeroftp-cli profiles --json

Structured JSON output: every command supports --json. Data goes to stdout, messages to stderr, so agents can pipe and parse cleanly.

# Output is machine-readable
aeroftp-cli stat --profile "Server" /var/www/index.html --json
# {"name":"index.html","size":4096,"type":"file","modified":"2026-03-15T10:30:00Z"}

# Pipe-friendly
aeroftp-cli cat --profile "Server" /config.ini | grep DB_HOST

Semantic exit codes: agents know exactly what failed without parsing text.

Code	Meaning
0	Success
1	Connection error
2	Not found
3	Permission denied
4	Transfer error
5	Configuration error
6	Authentication error
7	Not supported
8	Timeout

Batch scripting: agents can generate .aeroftp scripts for multi-step operations with error handling.

SET source=/var/www/html
ON_ERROR CONTINUE
CONNECT sftp://admin@server.com
GET {source}/database.sql ./backup/
GET {source}/uploads/ ./backup/uploads/
DISCONNECT

Clean output: NO_COLOR compliant, no ANSI escape sequences. Progress bars hidden when output is piped.

Compatible Agents and Frameworks

Agent / Framework	How it uses AeroFTP
Claude Code	Calls aeroftp CLI directly from terminal
Open Interpreter	Shell execution in its environment
Cline / Continue	Terminal tool calls in VS Code
Aider	Deployment after code changes
Cursor Agent	Shell command execution
Windsurf (Codium)	Terminal tool integration
Devin	Cloud-based agent with shell access
Codex CLI	OpenAI's CLI agent
Goose (Block)	Extensible agent with shell toolkit
SWE-agent	Software engineering research agent
AutoGPT / CrewAI	Multi-agent frameworks with shell tools
LangChain / LangGraph	Custom agents with subprocess tool
Computer Use (Anthropic, OpenAI Operator)	Desktop automation, can use both GUI and CLI
MCP servers	Model Context Protocol tools can wrap aeroftp commands

Real-World Scenarios

These are not hypothetical. Every command below works today with the AeroFTP CLI.

"Build my app and deploy it to the server"

An AI coding agent (Claude Code, Aider, Cline) makes code changes, builds the project, and deploys to production via FTP/SFTP. No manual file upload, no FTP client to open.

# Agent edits code, runs build, then deploys
npm run build
aeroftp-cli put --profile "Production" ./dist/ /var/www/html/ --json
# Verify deployment
aeroftp-cli ls --profile "Production" /var/www/html/ --json | jq '.entries | length'

"Back up my project to Google Drive every night"

A scheduled agent (cron + Claude Code, or AutoGPT) compresses a project and uploads it to cloud storage. The same profile-based flow works across FTP/SFTP, cloud drives, and object storage.

# Compress and upload to Google Drive
tar czf backup-$(date +%F).tar.gz ./project/
aeroftp-cli put --profile "Google Drive" ./backup-*.tar.gz /backups/
# Or sync an entire folder to S3
aeroftp-cli sync --profile "AWS S3" ~/documents/ /backup/documents/ --json
# Or to a NAS
aeroftp-cli sync --profile "NAS" ~/photos/ /media/photos/ --json

"Clone this repo, make changes, commit and push"

AeroFTP supports GitHub as a storage protocol. An agent can read files, create commits, and even open pull requests, all through the same CLI or AeroAgent tools.

# List repo contents
aeroftp-cli ls --profile "GitHub" /src/ --json
# Read a file
aeroftp-cli cat --profile "GitHub" /src/main.rs
# Upload a modified file (creates a commit)
aeroftp-cli put --profile "GitHub" ./main.rs /src/main.rs

Inside the GUI, AeroAgent can do the same through natural language: "read the README from my GitHub repo, update the version number, and commit it."

"Check how much space I have left across all my storage"

An agent audits storage usage across every connected service in seconds.

for profile in $(aeroftp-cli profiles --json | jq -r '.[].name'); do
  echo "=== $profile ==="
  aeroftp-cli df --profile "$profile" --json 2>/dev/null
done

Output for each profile includes used space, total space, and percentage. The agent can alert when any service is above 90%.

"Download all PDFs from Dropbox and re-upload them encrypted to S3"

Cross-provider file operations with encryption. The agent moves files between clouds, applying AeroVault encryption in between.

# Download from Dropbox
aeroftp-cli get --profile "Dropbox" "/contracts/*.pdf"
# Encrypt locally (AeroVault CLI or any tool)
# Re-upload to S3
aeroftp-cli put --profile "AWS S3" ./contracts/ /encrypted-archive/

"Set up a new developer's environment"

An onboarding agent provisions a developer's access to multiple servers and syncs starter configs.

# Pull configs from the team's shared WebDAV
aeroftp-cli get --profile "Team WebDAV" /dev-configs/ ./configs/
# Deploy SSH configs, editor settings, etc.
cp ./configs/.bashrc ~/
cp ./configs/.gitconfig ~/
# Pull project repos from GitHub
aeroftp-cli get --profile "GitHub" /src/ ./workspace/

"Monitor my servers and alert me"

A monitoring agent periodically checks connectivity and disk space.

# Test connectivity (exit code 0 = ok, 1 = down)
aeroftp-cli connect --profile "Production" --json
# Check if a critical file exists
aeroftp-cli stat --profile "Production" /var/www/html/index.html --json
# Find large log files
aeroftp-cli find --profile "Production" /var/log/ --name "*.log" --json

"Migrate files from one cloud to another"

Moving from Google Drive to a self-hosted Nextcloud (WebDAV). The agent handles the entire migration.

# List everything on Google Drive
aeroftp-cli ls --profile "Google Drive" / -r --json > inventory.json
# Download all files
aeroftp-cli get --profile "Google Drive" / ./migration/
# Upload to Nextcloud
aeroftp-cli put --profile "Nextcloud" ./migration/ / --json
# Verify
aeroftp-cli df --profile "Nextcloud" --json

AeroAgent: The Internal AI Assistant

For users who prefer the GUI, AeroAgent provides the same file management capabilities through natural language. It supports 19 AI providers and 47 tools covering:

• Local file ops: list, read, write, delete, rename, search, edit, mkdir, move, copy, batch rename, trash, file info, disk usage, find duplicates, grep, head, tail, stat, diff, tree
• Remote server ops: list, read, upload, download, delete, rename, mkdir, edit, search, batch upload/download, server exec
• System: shell execute, clipboard read/write, archive compress/decompress
• AI-specific: RAG index, RAG search, agent memory, edit preview

AeroAgent chains tool calls autonomously (up to 10 steps in normal mode, 50 in Extreme Mode) with a three-tier approval system based on tool danger level.

Full AeroAgent documentation →

CLI Command Reference

Core commands all support --json and --profile:

connect ls get put mkdir rm mv cat head tail touch hashsum check stat find df about dedupe tree sync batch profiles agent-info

All commands work identically across FTP, FTPS, SFTP, WebDAV, S3, Google Drive, Dropbox, OneDrive, MEGA, and all other supported protocols. The agent does not need to know which protocol a profile uses.

Full CLI documentation →

Design Principles

1. Humans first, agents welcome. The GUI remains the primary interface. Agent support is additive, not a compromise.
2. Credentials never exposed. Agents use profile names. Passwords stay in the encrypted vault.
3. Structured by default. --json is not an afterthought. Every command is designed for both human and machine consumption.
4. Protocol-agnostic. Agents do not need to know whether a profile points to FTP, SFTP, S3, or Google Drive. The same commands work everywhere.
5. Fail loudly. Semantic exit codes and stderr messages let agents handle errors programmatically.