A local content-addressed file store with compression and chunking.
This is Alpha level software.
casq is a single-binary tool for storing files and directories by their cryptographic hash. Think of it as a lightweight git object store or restic backend—but simpler, local-only, and purpose-built for content-addressed storage with modern efficiency features.
Why casq?
- Automatic deduplication - Identical content stored only once, even across different directories
- Content-defined chunking - Incremental backups with FastCDC v2020 (files ≥ 1MB split into variable chunks, 60-80% reuse after edits)
- Transparent compression - 3-5x storage reduction with zstd (files ≥ 4KB automatically compressed)
- Content addressing - Files identified by cryptographic hash, not by path (however, custom references are supported)
- Garbage collection - Reclaim space from unreferenced objects with mark & sweep
- Simple & fast - No databases, no network, just files on disk with BLAKE3 hashing
- Embeddable - Rust library (
casq_core) + CLI binary, easily integrated into your tools
# Use an environment variable to set the root. It is also possible to use --root/-R per invocation. Defaults to current working directory.
export CASQ_ROOT="/var/my-casq"
# Initialize the casq root - echoes the full path to the root directory.
casq initialize #>/var/my-casq
# Put a file system tree into the casq with a custom reference - echoes the hash
casq put /some/tree --reference tree@1 #>d856ec8e03cce04358fbd6a5135823574dea4de9ae6a6f511e3c060f33d144d4
# Put a blob into the casq with a custom reference via piping - echoes the hash
casq put --reference "big-blob@$(date -I)" - < curl https://big-blob.wherever #>d856ec8e03cce04358fbd6a5135823574dea4de9ae6a6f511e3c060f33d144d4
# Put a blob into the casq with a custom reference via piping - echoes the hash
curl https://example.org | casq put --reference "example.org@$(date +%s)" - #>454499efc25b742a1eaa37e1b2ec934638b05cef87b036235c087d54ee5dde59
# Get metadata about an object:
casq metadata tree@1 --json #>
#{
# "success": true,
# "result_code": 0,
# "type": "Tree",
# "hash": "d856ec8e03cce04358fbd6a5135823574dea4de9ae6a6f511e3c060f33d144d4",
# "entry_count": 1,
# "path": "./casq-store/objects/blake3-256/d8/56ec8e03cce04358fbd6a5135823574dea4de9ae6a6f511e3c060f33d144d4"
#}
# List the children of tree
casq list tree@1 #>file1\nfile2 etc
casq list --long tree@1 #>(listing of tree's children with most metadata)
# Read a blob and pipe it through less
casq get be616640346d71f639f55014a17229a8e6ab7d164b44a702d1c04da0d4bd48e3 | less
# List all references
casq references list
# Add a reference; key => value: casq references add REFERENCE HASH
casq references add my-custom-ref be616640346d71f639f55014a17229a8e6ab7d164b44a702d1c04da0d4bd48e3
# So now we can `get` it by ref
casq get my-custom-ref | less
# Remove the reference: casq references remove REFERENCE
casq references remove my-custom-ref
# Shows all objects that are unreferenced
casq find-orphans #>...
# Show all unreferenced data that would be deleted
casq collect-garbage --dry-run
# Deletes all unreferenced data
casq collect-garbageSo, general usage pattern:
initializeput(with --reference)- read:
metadata,get,list - manage references:
references add/list/removeandfind-orphans collect-garbage
Development
Run test coverage for the core crate with the mise task (requires cargo-tarpaulin):
# Generate HTML + XML coverage reports for casq_core
mise run coverage_core
# Reports are written to target/coverage/casq_core/Free, and open-source
Apache-2.0