pykx-dbapi

A PEP 249 (DB-API 2.0) wrapper around pykx, that gives marimo notebooks the ability to query kdb+/q servers.

It wraps a pykx.SyncQConnection in the standard connect() / Connection / Cursor interface, plus the ADBC extension methods marimo consumes: query results flow as Arrow (near zero-copy into polars), and marimo's datasources panel can browse tables and column types via q's tables[] and meta. marimo auto-detects the connection — no plugins needed (one optional config line fills the datasources panel; see Caveats). Queries are q/qSQL text sent to the server verbatim; there is no SQL translation.

pykx's unlicensed mode is sufficient — IPC-only use needs no kdb+ license on the client.

Screenshot

Adding to a marimo notebook project

uv-managed project (marimo installed in a project venv):

uv add pykx-dbapi
uv run marimo edit notebook.py

Plain pip:

pip install pykx-dbapi

Usage

In a marimo cell:

import pykx_dbapi

conn = pykx_dbapi.connect(host="localhost", port=5001)

marimo detects conn as a database connection. Create a SQL cell, pick conn as its engine, and write q or qSQL:

select avg price by sym from trades where size > 100

Results come back as dataframes. Non-table results are shaped into one: atoms/vectors as a single result column, dicts as key/value columns. Statements returning q's generic null (e.g. assignments like t:([] a:1 2 3)) produce no result set.

connect(...) passes its arguments straight to pykx.SyncQConnection, so TLS, auth, and timeouts all work: connect(host, port, username=..., password=...). You can also wrap an existing pykx connection: pykx_dbapi.Connection(existing).

Works outside marimo too — anything DB-API-aware can use it, e.g. pandas.read_sql("select from trades", conn) or polars.read_database(...). Errors raised by the wrapper follow the PEP 249 hierarchy (pykx_dbapi.Error and subclasses).

Arrow-only mode

By default every table result must convert to Arrow (arrow_only=True). Nested vector columns (a select-by without aggregation, e.g. 0!select date, close by ticker) convert to Arrow list columns — including temporals: nested dates land as list<date32>; months and minutes, which Arrow has no native unit for, are cast to timestamp[ms]/duration[ms].

Results pykx cannot convert raise pykx_dbapi.NotSupportedError telling you how to reshape — notably keyed tables (select ... by ...): unkey in the query with 0!select ....

connect(..., arrow_only=False) instead stitches keyed tables from their key/value parts and falls back to a Python-object conversion for tables pykx can't handle (slower; timestamps land at µs rather than ns precision).

q braces vs. marimo SQL cells

marimo compiles SQL cells to Python f-strings (that's how {python_var} interpolation works), so a literal {...} — a q lambda — is evaluated as a Python expression and typically fails with a NameError. Double the braces in SQL cells:

0!{{P:asc distinct x`ticker; exec P#(ticker!close) by date:date from x}} ungroup ...

Or run lambda-heavy queries from a Python cell, where no interpolation happens: mo.sql("0!{...} ...", engine=conn).

Caveats

• marimo treats kdb as a remote dialect and won't introspect eagerly. To populate the datasources panel automatically, enable discovery in marimo's config — in your project's pyproject.toml:

  [tool.marimo.datasources]
  auto_discover_schemas = true
  auto_discover_tables = true
  auto_discover_columns = true
  

  (or the same keys under [datasources] in ~/.config/marimo/marimo.toml to enable it globally).

• marimo's optional SQL-validation feature prefixes queries with EXPLAIN, which q rejects. Harmless — it surfaces as a validation warning; leave validation off for this connection.

• commit()/rollback() are no-ops: q IPC has no transactions.

• The deprecated q datetime (z) type is not supported (pykx refuses to convert it); cast to timestamp in your query, e.g. "p"$x.

• pykx's context interface is disabled by default (no_ctx=True): this wrapper never uses it, and its k)-dialect bootstrap breaks some q-only servers (e.g. KX Insights). Pass connect(..., no_ctx=False) to re-enable.

Development

uv sync
uv run pytest

Tests run against a fake pykx — no q server needed. They include contract tests that fail loudly if marimo's or polars' DB-API duck-typing ever changes.