akar-python — Python bindings (PyO3)

Python bindings for Akar — a pure Rust embedded graph database for AI agent memory.

Provides a drop-in replacement for KuzuDB in the Kairos project via a Cypher-to-SQL translation layer, or direct Akar API access.

Quick start

import akar

db = akar.Database("/path/to/db")
conn = akar.Connection(db)

# Execute Cypher queries
conn.query("""
    CREATE NODE TABLE Memory(id INT64, content STRING, embedding FLOAT[],
    PRIMARY KEY(id))
""")

# Parameterized queries
result = conn.execute(
    "MATCH (m:Memory {id: $id}) RETURN m.content",
    {"id": 42},
)

while result.has_next():
    print(result.get_next())

# Get all results as list of dicts
rows = result.get_all()

db.close()

API surface

Database(path: str)

• close() — release file lock, drop all connections
• __repr__() — display path

Connection(database: Database)

• query(cypher: str) -> QueryResult — execute Cypher, no parameters
• execute(cypher: str, params: dict = None) -> QueryResult — execute Cypher with $param interpolation
• close() — drop connection (DB stays open)

QueryResult

• has_next() -> bool — check for more rows
• get_next() -> dict | list | scalar — fetch next row (returns dict when rows_as_dict=True, default)
• get_all() -> list — fetch all rows
• get_column_names() -> list[str] — column names
• rows_as_dict(state: bool) — toggle dict vs tuple mode
• __iter__() / __len__() / __bool__() — iteration and checks
• close() — drop result

Kairos compatibility shim

For drop-in replacement of KuzuDB in Kairos, use the shim at kairos/kuzu.py:

# In Kairos codebase:
import kuzu  # transparently becomes `import akar`

db = kuzu.Database("/path/to/db")
conn = kuzu.Connection(db)

The shim registers a sys.modules alias for the kuzu package (ladybug alias was removed in 0.1.1 — Kairos imports akar directly).

Supported Cypher features

• DDL: CREATE NODE TABLE, CREATE REL TABLE, DROP TABLE, ALTER TABLE, CREATE/DROP INDEX
• DML: MATCH, WHERE, RETURN, CREATE, SET, DELETE, MERGE (ON CREATE/ON MATCH)
• Composition: WITH, ORDER BY, LIMIT, SKIP, UNION ALL, UNWIND, OPTIONAL MATCH, FOREACH
• Data loading: COPY FROM (CSV, Parquet), COPY TO, EXPORT DATABASE, IMPORT DATABASE
• Parameters: $param interpolation (string, int, float, bool, list, dict, null)
• Translation: Kuzu syntax auto-converted to Akar SQL (IF NOT EXISTS, FLOAT[n] → FLOAT[], INSTALL/LOAD EXTENSION no-op, vector index calls)

Build from source

This crate is a standalone workspace (not a member of akar-core's workspace), so it does not affect the test [akar-core] gate or CI.

Prerequisites

• Rust toolchain (stable)
• Python >= 3.9
• maturin >= 1.7

Development build

cd akar-core/akar-python

python -m venv .venv
.venv/Scripts/python -m pip install maturin pytest
$env:PYO3_PYTHON = ".venv\Scripts\python.exe"   # Windows PowerShell
# export PYO3_PYTHON=".venv/bin/python"           # Linux/macOS

.venv/Scripts/maturin develop    # build + install into venv
.venv/Scripts/python -c "import akar; print(akar.__name__)"

Release build (wheel)

.venv/Scripts/maturin build --release
# Output: target/wheels/akar-0.1.0-cp39-abi3-<platform>.whl

Source distribution

.venv/Scripts/maturin sdist
# Output: target/wheels/akar-0.1.0.tar.gz

Publishing to PyPI

akar 0.1.3 is live on PyPI (wheel cp39-abi3 + sdist, published 2026-08-24). Procedure:

1. Build artifacts (do NOT use maturin publish — it rebuilds the whole workspace from scratch, ~10 min, and can exceed tool timeouts):

   .venv/Scripts/maturin build --release
   .venv/Scripts/maturin sdist
   

2. Upload with twine (installed via uv tool install twine). Create an API token at https://pypi.org/manage/account/token/ ("Entire account"); it shows once — store it in a file outside the repo.

   # TestPyPI first (optional but recommended)
   twine upload --repository testpypi target/wheels/*.whl target/wheels/*.tar.gz
   
   # PyPI production
   twine upload --repository pypi target/wheels/*.whl target/wheels/*.tar.gz
   

   Token supplied via env: TWINE_USERNAME=__token__, TWINE_PASSWORD=<token>. Add --skip-existing when re-uploading after a metadata-only change.
3. Verify: pip install akar==<version> in a clean venv, then smoke test Database/Connection/Cypher. Run the compat harness (tests/test_kuzu_compat.py) against the installed wheel.
4. Bump version in pyproject.toml ([project].version) before each new release — PyPI rejects re-upload of an existing version.
5. Record the release in ../CHANGELOG.md under ## [Unreleased] (Keep a Changelog).

Tests

Rust unit tests (cargo)

cd akar-core/akar-python
cargo test --lib

39 tests covering: translation layer, parameter interpolation, value conversion, schema bootstrap, vector index / kNN, community detection (Louvain), LSTM bindings, spread activation, lock management, connection lifecycle.

Python compatibility harness (pytest)

cd akar-core/akar-python
.venv/Scripts/pytest tests/test_kuzu_compat.py -v

53 tests validating drop-in replacement against real Kairos KuzuDBStore / KuzuDBDreamBackend methods (schema bootstrap, store, search, connections, export/repair, CHECKPOINT).

Project structure

akar-core/akar-python/
├── Cargo.toml          # pyo3 0.29.2 (abi3-py39), standalone workspace
├── pyproject.toml      # maturin build config
├── src/
│   ├── lib.rs          # PyO3 module: Database, Connection, QueryResult
│   ├── translation.rs  # Kuzu Cypher → Akar SQL translation
│   └── param_interp.rs # $param → literal interpolation
└── tests/
    ├── conftest.py     # pytest fixtures + gap-report hooks
    └── test_kuzu_compat.py  # 53 Kairos drop-in tests

License

GPL-3.0-or-later