turingdb

name

turingdb

description

Start, query, write, and manage TuringDB columnar graph databases using the Python SDK and Cypher dialect. TRIGGER when: code imports `turingdb` or `TuringDB`; user mentions TuringDB, turing db, or turing database; user asks about TuringDB Cypher queries, graph versioning with changes/commits, vector search in TuringDB, or the `turingdb` CLI; files contain TuringDB connection strings (localhost:6666) or TuringDB SDK calls (`client.query`, `client.new_change`, `client.load_graph`). SKIP: generic graph/Neo4j/Cypher questions with no TuringDB mention; general database work unrelated to TuringDB.

TuringDB

TuringDB is a columnar graph database with git-like versioning. Its query language is a Cypher dialect — largely compatible with standard Cypher but with TuringDB-specific extensions. The Python SDK query() method returns results as pandas DataFrames.

Setup

Default: connect to a running TuringDB server over HTTP. The server listens on http://localhost:6666 by default; start one with the turingdb CLI if it isn't already running (see startup.md).

from turingdb import TuringDB

client = TuringDB(host="http://localhost:6666")   # HTTP/JSON client (the default backend)
client.create_graph("my_graph")  # create new graph
# or
client.load_graph("my_graph")    # load an existing on-disk graph by name
client.set_graph("my_graph")     # make it active

TuringDB(...) supports three backends via type: "json" (HTTP, the default — pass host="http://localhost:6666"), "native" (binary protocol over HTTP — host=/port=), and "embedded" (in-process, no server — pass data_dir=). Use the server (json) by default; it's required for the browser visualizer (-ui), list_available_graphs() to discover on-disk graphs, concurrent multi-process access, and S3 transfers. Reach for "embedded" only when you want a self-contained, in-process engine with no server to manage. See startup.md for connecting/starting the server and introspection.md for the full backend reference.

Routing

Based on what the user is asking, immediately read the matching file from this same directory using the Read tool. Do not ask the user which file to read — determine it from context.

Task	File
Connect to a TuringDB server (or run embedded) and load/create a graph	startup.md
Reading data — MATCH, WHERE, filtering, traversal, joins	querying.md
Writing data — CREATE, SET, updating the graph	writing.md
Importing external data — JSONL, GML, Parquet, Neo4j migration	importing.md
Graph algorithms — shortest path, vector/embedding search	algorithms.md
Exploring an unfamiliar graph, versioning, time travel, GML/JSONL import	introspection.md
Importing Parquet files via the turing-parquet CLI (turingdb ≥ 1.32)	parquet.md

If the task spans multiple areas (e.g. connect then query), read the files in sequence.