A Python loader for doltlite — Dolt's version control through SQLite's drop-in API.

import doltlite          # one-time bootstrap
import sqlite3

conn = sqlite3.connect("repo.db")
conn.execute("CREATE TABLE t(x INT)")
conn.execute("INSERT INTO t VALUES (1)")
conn.execute("SELECT dolt_commit('-A', '-m', 'init')")

import doltlite makes libdoltlite the active SQLite engine for the current interpreter, so anything that uses the standard sqlite3 module — including SQLAlchemy with sqlite:///... URLs — transparently gains dolt_commit, dolt_branch, dolt_merge, dolt_log, dolt_diff_<table>, and the other Dolt SQL functions and virtual tables.

pip install doltlite

Wheels bundle a precompiled libdoltlite for macOS (arm64) and Linux (x86_64, aarch64). No system-level setup required.

Intel Mac wheels aren't shipped in v0.11.x — GitHub's free-tier macos-13 runners queue for hours and block releases. Intel Mac users should build libdoltlite locally and use the DOLTLITE_LIB path below.

For development against a local checkout of doltlite, point DOLTLITE_LIB at your built library instead:

DOLTLITE_LIB=/path/to/libdoltlite.dylib python3 your_script.py

The package piggybacks on Python's stdlib sqlite3, which must load SQLite as a shared extension at runtime. The following Pythons work:

• Distro / system Python (Linux)
• Homebrew Python (macOS, Linux)
• pyenv-built Python
• Conda Python

These do not work because their _sqlite3 is statically linked into the interpreter:

• python-build-standalone interpreters — the default for uv python install, mise, and Rye

If you use uv, target one of the supported Pythons explicitly:

uv venv --python /opt/homebrew/bin/python3   # or /usr/bin/python3

Doltlite is a SQLite drop-in: it implements the same sqlite3_* C API and adds Dolt-specific functions and virtual tables on top. To use it from Python, libdoltlite has to be loaded ahead of the system libsqlite3 so its symbols win during the dynamic-link symbol-resolution pass that Python's _sqlite3 module triggers when it's first imported.

The mechanics differ by platform:

ELF flat-namespace symbol resolution: if libdoltlite is loaded into the process with RTLD_GLOBAL before _sqlite3 is loaded, its sqlite3_* symbols enter the global namespace and the later libsqlite3.so lookup finds them first.

import doltlite does ctypes.CDLL(libdoltlite_path, mode=ctypes.RTLD_GLOBAL) in this case — no re-exec required.

If sqlite3 was already imported before doltlite, that boat has sailed: we fall back to re-execing the interpreter with LD_PRELOAD set, so libdoltlite is loaded at process start.

macOS uses a two-level namespace: _sqlite3.so has an LC_LOAD_DYLIB command bound to a specific libsqlite3.dylib path (e.g. /opt/homebrew/opt/sqlite/lib/libsqlite3.dylib). Plain dlopen / ctypes.CDLL does not redirect that resolution, and DYLD_INSERT_LIBRARIES alone doesn't either — the inserted library has to have an install_name (LC_ID_DYLIB) that matches the path _sqlite3.so was linked against.

import doltlite does:

1. Detect that path via otool -L $(python3 -c 'import _sqlite3; print(_sqlite3.__file__)').
2. Copy libdoltlite to $TMPDIR/.../libsqlite3.dylib (cached per (lib, mtime, install_name)).
3. Rewrite the shim's install_name with install_name_tool -id <path>.
4. Re-exec the interpreter with DYLD_INSERT_LIBRARIES=<shim>.

The two-level lookup then accepts the shim because the install_name matches.

This requires otool and install_name_tool — install Xcode Command Line Tools (xcode-select --install) if missing.

When a re-exec is required (macOS, or Linux-after-sqlite3-loaded), the bootstrap calls os.execvpe with sys.argv. That means the invocation must name a script file we can replay:

• ✅ python3 my_script.py
• ✅ python3 -m my_package
• ❌ python3 -c "import doltlite; ..." — code string isn't in argv
• ❌ Interactive REPL / Jupyter — there's no script to re-exec

In the unsupported cases the bootstrap raises DoltliteLoadError with a clear workaround: set DYLD_INSERT_LIBRARIES (macOS) or LD_PRELOAD (Linux) yourself before starting Python.

import doltlite

# Side-effect import does the bootstrap automatically. Subsequent
# imports are no-ops thanks to a process-env marker.

# If you want to bootstrap explicitly (e.g. inside a function):
doltlite.bootstrap()

# Find where the loaded libdoltlite came from:
doltlite.libdoltlite_path()  # absolute path

• doltlite — the underlying SQLite + Dolt engine
• doltlite-sqlalchemy-getting-started — end-to-end SQLAlchemy demo (commits, branches, schema change, merge)

Apache License 2.0. See LICENSE.