Manage and sync files across 70+ cloud storage platforms

Rclone CLI is a reference skill for managing cloud files from the terminal. It covers uploads, downloads, copies, moves, synchronization, remote configuration, mounts, listings, and integrity checks. Its command vocabulary provides cloud equivalents to familiar file utilities, while remote paths identify a configured storage service and a path within it. Local directories can participate in the same transfer workflows. The skill is licensed under Apache-2.0.

The guide distinguishes operations that are easy to confuse. Copy transfers files without deleting extra destination content. Sync makes the destination match the origin and propagates deletions. Move transfers files and then removes the originals. Check compares source and destination, while cryptcheck handles comparison through an encrypted remote. Timestamp preservation and checksum verification depend on backend support and the selected transfer options. Transfers can restart from the last good file, rather than promising recovery at an arbitrary byte offset.

Alongside the main command reference, the material includes detailed filtering, encryption, cache, chunker, union, and combine references. These explain how to select files, wrap existing storage with additional behavior, and present multiple storage locations through a common view.

Use this skill for terminal-based cloud file operations, including local-to-cloud mirrors, cloud-to-cloud migrations, restores, bandwidth-limited transfers, and encrypted mirrors. A scheduled sync is a mirror, not an independent versioned backup: unwanted changes and deletions can propagate to the destination. Keep independent versioned backups where recovery history matters.

The executable workflow requires an already installed and reviewed rclone binary. Start by checking it:

rclone --version

If the binary is missing, the guide requires stopping that workflow. Installation is a separate task involving an immutable signed release, checksum or signature verification, inspection, and approval. The bundled historical installer is disabled. The guide also calls for checking commands, dependencies, credentials, and service behavior against the actual environment before making changes.

Preview sync, move, delete, and purge operations with dry-run before execution. Interactive mode adds confirmation for destructive operations. Purge ignores filters and removes everything beneath the selected path. A writable mount can lose data when interrupted; full VFS caching needs sufficient local disk and successful upload and flush, and does not guarantee durability. The deprecated cache backend is discouraged for new installations and has documented bugs and throttling risks.

Inputs include local paths, configured remote paths, operation choices, filter rules, and transfer settings. The reference covers configuration creation and updates, listing remotes, quota inspection, progress, logging, bandwidth limits, transfer concurrency, and checksum-based comparisons. Outputs depend on the selected command: transferred files, synchronized directories, listings, integrity results, a mounted filesystem, or a file-serving endpoint.

File selection can use path patterns, regular expressions, file lists, size and age thresholds, directory marker files, and metadata patterns. Rules use first-match processing; include rules imply a final exclusion, while a positive filter rule alone does not. Prefer filter rules when combining inclusion and exclusion instead of mixing flag families. File-list flags override other filters, and the raw variant preserves whitespace and comment-like names. Quote shell metacharacters and values containing spaces. Debug logging and filter dumps help inspect the resulting selection before modifying files.

Hash filtering selects deterministic subsets for partitioned work or sampled integrity checks. It still traverses directories and must not be combined with delete-excluded, which could remove unselected destination files. Directory filters can reduce traversal, but regular-expression filters disable that optimization. Metadata filters operate on object attributes rather than names.

Crypt wraps an existing remote or a local directory and encrypts and decrypts data on the client. Accessing the underlying storage directly bypasses that translation. File and directory name encryption are configurable; filename obfuscation is weak protection, and encrypted names can encounter provider path limits. Retain the password and any custom salt securely. Changing a configured password does not re-encrypt existing files: migration requires decrypting and uploading through a newly configured crypt remote. Crypt integrity checking and modification-time support follow different rules from ordinary checksum comparisons.

Chunker splits large files for providers with file-size limits and assembles them on download. Its metadata, naming, hash, and transaction settings affect compatibility and recovery. It needs server-side move or copy-and-delete support, and temporary-file handling can add overhead for small files. Interrupted work can leave hidden chunks consuming quota. Changing critical chunking settings requires migration rather than editing a live configuration and expecting existing chunks to change automatically.

Provider references cover Amazon S3 and compatible storage, Google Drive, Dropbox, Microsoft OneDrive, Azure Blob Storage, Backblaze B2, Google Cloud Storage, SFTP, WebDAV, OpenStack Swift, and FTP. Provider capabilities determine which comparison and transfer behaviors are available.

Serving modes include HTTP, WebDAV, SFTP, FTP, S3, DLNA, a Restic backend, and a Docker volume plugin API. The Docker mode is not a container registry. Remote control should bind to loopback; deliberate remote access requires authentication, TLS, and network restrictions.

Union overlays several upstreams with separate action, creation, and search policies. Existing-path policies preserve placement, quota-based policies need supporting quota fields, and read-only or no-create tags constrain writes. Its writeback option can copy remote reads into a local upstream, but cache expiry and size management remain the user's responsibility. Combine instead places upstreams beneath named directories in a synthetic tree, including configured Google Drive Shared Drives.

The legacy cache reference covers read buffering, queued offline uploads, Plex playback detection, remote-control operations, and configuration tradeoffs. It also warns about directory-cache freshness, experimental Windows mounting, provider throttling, and cache/crypt ordering. Absolute and relative SFTP paths can produce different cache keys, so keep path forms consistent. This is separate from VFS caching. Reference material also points to bisync, hasher, installation, Docker usage, provider comparisons, and individual command documentation.

Protect the rclone configuration file and restrict its permissions: it can contain passwords, private keys, and OAuth tokens. Default password obscuring is reversible, not secure encryption. A redacted configuration still needs inspection before sharing, and credentials must not be committed or logged. These precautions apply across all configured providers and overlays.