vost 0.66.1

A versioned filesystem backed by bare Git repositories

vost

vost is a versioned filesystem backed by bare Git repositories. Store, retrieve, and version directory trees of files with text and binary data using an immutable-snapshot API. Unlike Git, every write (or batch of writes and deletes) produces a new commit. Old snapshots remain accessible forever.

The repositories are standard Git repositories that can be manipulated with Git tools as well.

vost includes an intuitive Python API and an optional command line interface. The CLI includes operations to create repositories, copy and write into or out of them using rsync-like syntax, archive to zip or tar files, even act as an HTTP or Git server for a given snapshot or branch.

Installation

pip install vost            # core library (dulwich only)
pip install "vost[cli]"    # adds the vost command-line tool

Or run the CLI without installing:

uvx "vost[cli]" -r myrepo.git ls : # run via uvx
uv tool install "vost[cli]"       # or install as a persistent tool

Requires Python 3.10+.

Note: The gc and backup/restore CLI commands shell out to an installed git executable. All other commands and the entire Python API are self-contained (pure dulwich).

Quick start

from vost import GitStore

# Create (or open) a repository with a "main" branch
repo = GitStore.open("data.git")

# Get a snapshot of the current branch ("main" by default)
fs = repo.branches.current

# Write a file -- returns a new immutable snapshot
fs = fs.write_text("hello.txt", "Hello, world!")

# Read it back
print(fs.read_text("hello.txt"))  # 'Hello, world!'

# Every write is a commit
print(fs.commit_hash)           # full 40-char SHA
print(fs.message)               # '+ hello.txt'

Core concepts

Bare repository. vost uses a bare Git repository -- one that contains only Git's internal object database, with no working directory or checked-out files. You won't see your stored files by browsing the repo directory; all data lives inside Git's content-addressable object store and is accessed exclusively through the vost API. This is by design: it avoids filesystem conflicts, keeps the storage compact, and lets Git handle deduplication and integrity.

GitStore opens or creates a bare repository. It exposes branches and tags as MutableMapping objects (supporting .get, .keys, .values, .items, etc.).

FS is an immutable snapshot of a committed tree. Reading methods (read, ls, walk, exists, open) never mutate state. Writing methods (write, write_from_file, remove, batch) return a new FS pointing at the new commit -- the original FS is unchanged.

Snapshots obtained from branches are writable (fs.writable == True). Snapshots obtained from tags are read-only (fs.writable == False).

API

Opening a repository

repo = GitStore.open("data.git")                         # create or open (default branch: "main")
repo = GitStore.open("data.git", create=False)            # open only
repo = GitStore.open("data.git", branch="dev")            # custom default branch
repo = GitStore.open("data.git", branch=None)             # branchless
repo = GitStore.open("data.git", author="alice",          # custom author
                     email="[e-mail]")

Branches and tags

fs = repo.branches["main"]
repo.branches["experiment"] = fs   # fork a branch
del repo.branches["experiment"]    # delete a branch

repo.tags["v1.0"] = fs            # create a tag
snapshot = repo.tags["v1.0"]       # read-only FS

repo.branches.current_name           # "main"
fs = repo.branches.current           # FS for the current branch
repo.branches.current = "dev"        # set current branch

for name in repo.branches:
    print(name)
"main" in repo.branches           # True

Reading

data = fs.read("path/to/file.bin")           # bytes
text = fs.read_text("config.json")           # str (UTF-8)
chunk = fs.read("big.bin", offset=100, size=50)  # partial read (50 bytes at offset 100)
chunk = fs.read_by_hash(sha, offset=0, size=1024)  # read blob by SHA, bypasses tree walk

entries = fs.ls()                             # root listing — list of name strings
entries = fs.ls("src")                        # subdirectory listing
details = fs.listdir("src")                  # list of WalkEntry (name, oid, mode)
exists = fs.exists("path/to/file.bin")        # bool
info = fs.stat("path/to/file.bin")           # StatResult (mode, file_type, size, hash, nlink, mtime)
ftype = fs.file_type("run.sh")               # FileType.EXECUTABLE
nbytes = fs.size("path/to/file.bin")         # int (bytes)
sha = fs.object_hash("path/to/file.bin")     # 40-char hex SHA
tree_sha = fs.tree_hash                      # root tree 40-char hex SHA

# Walk the tree (like os.walk)
for dirpath, dirnames, file_entries in fs.walk():
    for entry in file_entries:
        print(entry.name, entry.file_type)    # WalkEntry with name, oid, mode

# Glob
matches = fs.glob("**/*.py")                 # sorted list of matching paths

# Partial read (offset + size)
header = fs.read("data.bin", offset=0, size=4)

Writing

Every write auto-commits and returns a new snapshot:

from vost import FileType

fs = fs.write_text("config.json", '{"key": "value"}')
fs = fs.write_text("script.sh", "#!/bin/sh\n", mode=FileType.EXECUTABLE)
fs = fs.write_text("config.json", "{}", message="Reset")   # custom commit message
fs = fs.write("image.png", raw_bytes)                       # binary data
fs = fs.write_from_file("big.bin", "/data/big.bin")         # from disk
fs = fs.write_symlink("link", "target")                     # symlink
fs = fs.remove("old-file.txt")

# Buffered write (commits on close)
with fs.writer("big.bin") as f:
    f.write(chunk1)
    f.write(chunk2)
fs = f.fs

# Text mode
with fs.writer("log.txt", "w") as f:
    f.write("line 1\n")
    f.write("line 2\n")
fs = f.fs

# Inside a batch
with fs.batch() as b:
    with b.writer("streamed.bin") as f:
        for chunk in source:
            f.write(chunk)

The original FS is never mutated:

fs1 = repo.branches["main"]
fs2 = fs1.write("new.txt", b"data")
assert not fs1.exists("new.txt")  # fs1 is unchanged
assert fs2.exists("new.txt")

Batch writes

Multiple writes/removes in a single commit:

with fs.batch(message="Import dataset v2") as b:
    b.write("a.txt", b"alpha")
    b.write_from_file("big.bin", "/data/big.bin")
    b.write_symlink("link.txt", "a.txt")
    b.remove("old.txt")
fs = b.fs  # new snapshot after the batch commits

If an exception occurs inside the batch, nothing is committed.

History

parent = fs.parent                               # FS or None
ancestor = fs.back(3)                            # 3 commits back

for snapshot in fs.log():                        # full commit log
    print(snapshot.commit_hash, snapshot.message)

for snapshot in fs.log("config.json"):           # file history
    print(snapshot.commit_hash, snapshot.message)

for snapshot in fs.log(match="deploy*"):         # message filter
    ...

for snapshot in fs.log(before=cutoff):           # date filter
    ...

fs = fs.undo()                                   # move branch back 1 commit
fs = fs.redo()                                   # move branch forward 1 reflog step

# Reflog — branch movement history
for entry in repo.branches.reflog("main"):
    print(entry.old_sha, entry.new_sha, entry.message)

Copy and sync

# Disk to repo (current branch)
fs = fs.copy_in(["./data/"], "backup")
print(fs.changes.add)                            # [FileEntry(...), ...]

# Repo to disk
fs.copy_out(["docs"], "./local-docs")

# Work with a non-default branch
dev = repo.branches["dev"]
dev = dev.copy_in(["./features/"], "src")

# Copy between branches (atomic, no disk I/O)
main = repo.branches["main"]
dev = dev.copy_from_ref(main, "config")               # dir mode: config/ → config/
dev = dev.copy_from_ref(main, "config/", "imported")  # contents mode: config/* → imported/
dev = dev.copy_from_ref(main, "config", "imported")   # dir mode: config/ → imported/config/

# Sync (make identical, including deletes)
fs = fs.sync_in("./local", "data")
fs.sync_out("data", "./local")

# Expand globs on disk (same dotfile rules as fs.glob)
from vost import disk_glob
files = disk_glob("./data/**/*.csv")

# Remove and move within repo
fs = fs.remove(["old-dir"], recursive=True)
fs = fs.move(["old.txt"], "new.txt")

Atomic apply

Apply multiple writes and removes in a single commit without a context manager:

from vost import WriteEntry

fs = fs.apply(
    writes={
        "config.json": b'{"v": 2}',
        "script.sh": WriteEntry(data=b"#!/bin/sh\n", mode=0o100755),
        "link": WriteEntry(target="config.json"),          # symlink
    },
    removes=["old.txt", "deprecated/"],
    message="Update config and clean up",
)

Snapshot properties

fs.commit_hash           # str -- full 40-character commit SHA
fs.ref_name              # str | None -- ref name (branch or tag), or None for detached
fs.message               # str -- commit message
fs.time                  # datetime -- commit timestamp (timezone-aware)
fs.author_name           # str -- commit author name
fs.author_email          # str -- commit author email
fs.changes               # ChangeReport | None -- changes from last operation

Backup and restore

diff = repo.backup("https://github.com/user/repo.git")    # MirrorDiff
diff = repo.restore("https://github.com/user/repo.git")   # MirrorDiff
diff = repo.backup(url, dry_run=True)                      # preview only

Concurrency safety

vost uses an advisory file lock (vost.lock in the repo directory) to make the stale-snapshot check and ref update atomic on a single machine. If a branch advances after you obtain a snapshot, attempting to write from the stale snapshot raises StaleSnapshotError:

from vost import StaleSnapshotError

fs = repo.branches["main"]
_ = fs.write("a.txt", b"a")     # advances the branch

try:
    fs.write("b.txt", b"b")     # fs is now stale
except StaleSnapshotError:
    fs = repo.branches["main"]  # re-fetch and retry

For single-file writes, retry_write handles the re-fetch-and-retry loop automatically with exponential backoff:

from vost import retry_write
fs = retry_write(repo, "main", "file.txt", data)

Guarantees and limitations:

• Single-machine, multi-process writes to the same branch are serialized by the file lock and will never silently lose commits.
• When a stale write is rejected, the commit object is created but unreferenced. These dangling objects are harmless and will be cleaned up by git gc.
• Cross-machine coordination (e.g. NFS-mounted repos) is not supported -- file locks are not reliable over network filesystems.

Maintenance: vost repos are standard bare Git repositories. Run vost gc (or git gc directly) to repack loose objects and prune unreferenced data. This is optional but can reduce disk usage for long-lived repos.

Error handling

Exception	When
FileNotFoundError	read/remove on a missing path; write_from_file with a missing local file; opening a missing repo with create=False
IsADirectoryError	read on a directory path; write_from_file with a directory; remove on a directory
NotADirectoryError	ls/walk on a file path
PermissionError	Writing to a tag snapshot
KeyError	Accessing a missing branch/tag; overwriting an existing tag
ValueError	Invalid path (.., empty segments); unsupported open mode
TypeError	Assigning a non-FS value to a branch or tag
RuntimeError	Writing/removing on a closed Batch
StaleSnapshotError	Writing from a snapshot whose branch has moved forward

CLI

vost includes a command-line interface. Install with pip install "vost[cli]" or uv tool install "vost[cli]".

export VOST_REPO=/path/to/repo.git    # or pass --repo/-r per command

Repo paths and the : prefix

Because vost commands work with both local files and files stored in the repo, you need a way to tell them apart. A leading : marks a repo path. Without it, the argument is a local filesystem path.

:file.txt              repo path on the current branch
:                      repo root
main:file.txt          repo path on the "main" branch
v1.0:data/             repo path on the "v1.0" tag
main~3:file.txt        3 commits back on main

This applies to cp, sync, rm, mv, ls, cat, and other commands. For ls, cat, rm, and write the : is optional (arguments are always repo paths), but it is required for cp, sync, and mv to distinguish repo paths from local paths.

For full details on path parsing, ancestor syntax (~N), and interaction with flags, see Path Syntax.

# Repository management
vost init
vost destroy -f
vost gc

# Copy files (disk <-> repo, repo <-> repo)
vost cp local-file.txt :                        # disk to repo root
vost cp ./mydir :dest                            # copy mydir into dest/mydir
vost cp ./mydir/ :dest                           # trailing / = contents only
vost cp '/data/./logs/app' :backup               # /./  pivot: → backup/logs/app/...
vost cp './src/*.py' :backup                     # glob
vost cp :file.txt ./local.txt                    # repo to disk
vost cp -n ./mydir :dest                         # dry run

# Sync (make identical, including deletes)
vost sync ./local :repo_path
vost sync :repo_path ./local
vost sync --watch ./dir :data                    # continuous watch mode

# Browse
vost ls
vost ls -R :src
vost cat file.txt

# Write stdin
echo "hello" | vost write file.txt
cmd | vost write log.txt -p | grep error         # passthrough (tee)

# Remove and move within repo
vost rm old-file.txt
vost rm -R :dir
vost mv :old.txt :new.txt
vost mv ':*.txt' :archive/

# History
vost log
vost log --path file.txt --format jsonl
vost diff --back 3
vost undo
vost redo

# Branches and tags
vost branch set dev --ref main
vost branch exists dev
vost tag set v1.0
vost tag delete v1.0

# Archives
vost archive_out out.zip
vost archive_in data.tar.gz

# Mirror (backup/restore all refs)
vost backup https://github.com/user/repo.git
vost restore https://github.com/user/repo.git
vost backup -n https://github.com/user/repo.git  # dry run

# Serve files over HTTP
vost serve                                        # single branch
vost serve --all --cors                           # all refs with CORS

# Serve repo over Git protocol (read-only)
vost gitserve

For full CLI documentation, see CLI Reference.

Git notes

Attach metadata to commits without modifying history:

# Default namespace (refs/notes/commits)
ns = repo.notes.commits
ns[fs.commit_hash] = "reviewed by Alice"
print(ns[fs.commit_hash])                       # "reviewed by Alice"
del ns[fs.commit_hash]

# Custom namespaces
reviews = repo.notes["reviews"]
reviews[fs.commit_hash] = "LGTM"

# Shortcut: note for the current HEAD commit
ns.for_current_branch = "deployed to staging"
print(ns.for_current_branch)

# Batch writes (single commit)
with repo.notes.commits.batch() as b:
    b[hash1] = "note one"
    b[hash2] = "note two"

# Iteration
for commit_hash, text in ns.items():
    print(commit_hash, text)

Documentation

• Documentation hub -- quick start and navigation
• Python API Reference -- classes, methods, and data types
• CLI Reference -- the vost command-line tool
• Path Syntax -- how ref:path works across commands
• GitHub Repository -- source code, issues, and releases

Development

uv sync --dev       # install with dev dependencies (includes CLI)
uv run python -m pytest -v