Author of #chDB (acq. by ClickHouse)
Singapore
Joined February 2009
- Tweets1.2K
- Following472
- Followers2.2K
- Likes663
No GIL, Real Gains: Porting a C++ Python Extension to Free-Threaded Python 3.14 @PyData Amsterdam 2026 youtu.be/soBWS_SlqJc?si=LQ5_…
auxten retweeted
The latest pg_clickhouse and chdb extension releases, out today, greatly expand Postgres/ClickHouse data type mapping, including intervals and composite types, and provide tools to handle character encoding issues.
buff.ly/Fg365fE
Can someone tell me how I'm supposed to use the $250 cloud environment credit that Claude Code gives you?
Most of my projects are developed locally. I've got a lot of set-up debugging environments locally, along with various environment variables, keys, emulators, and large test data that can't be committed to Git (like a bunch of parquet files). I really can't figure out how to make use of those $250 tokens.
How is everyone else using them? Any inspiration or ideas would be appreciated.
My Claude Code session hasn't written code in weeks. It's a director: every task goes to a local CLI worker in tmux, tracked by an on-disk ledger.
New: who gets the task and at what effort is decided by @typesafeai Jev — a decision model, not an LLM. 6 typed answers, ~1s.
auxten retweeted
Nice writeup: instead of chunking a too-big Parquet, swap `import pandas as pd` for `import chdb.datastore as pd` and keep the same DataFrame code. Lazy and columnar underneath, so reading 3 of 40 columns never pays for the other 37.
dev.to/herbze/when-pandas-ru…
Meet the creator of @DataPolars @RitchieVink at PyData AMS. I am amazed that a company of only 20+ people built such a great data engine.
Python 3.14 can finally run Python threads in parallel — but native extensions don’t get the speedup for free.
At PyData Amsterdam this Friday, I’ll share what it took to make a C++-backed Python extension actually benefit from free-threading: what broke, what surprised us, and what the benchmarks look like on 16 cores.
No GIL. Real gains. 🐍⚡
See you at 14:10 in Entropy!
🤖 Made with AI
I'll be speaking at PyData Amsterdam 2026 about how we used Python's new thread-free features to boost chDB UDF performance by 10x. Let's connect at the venue, or feel free to reach out for a coffee.
pretalx.com/pydata-amsterdam…
I'll be speaking at PyData Amsterdam 2026 about how we used Python's new thread-free features to boost chDB UDF performance by 10x. Let's connect at the venue, or feel free to reach out for a coffee.
pretalx.com/pydata-amsterdam…
When I started chDB, the bet was simple: ClickHouse’s 70+ formats and object-storage stack should be reusable, not reimplemented for every database.
That’s now a Postgres extension: chdb + chdb_hook.
COPY FROM 's3://...'
CREATE TABLE ... WITH (copy_from = '...')
vs pg_duckdb / pg_lake: more consistent imports.
vs aws_s3: not even close on format coverage.
Design choice that matters: the engine is not linked into the backend. A helper starts, talks over fds, exits. Occasional ETL should not pin a full OLAP engine in shared memory.
Post: clickhouse.com/blog/introduc…
Ext: pgxn.org/dist/chdb/
New: chDB as a Postgres extension.
Import/export from S3, GCS, Azure, HTTP, and local files — in the formats ClickHouse already supports (Parquet, Arrow, ORC, JSON, Avro, and more).
COPY and CREATE TABLE ... copy_from just work. Helper process on demand, not an engine living inside Postgres.
clickhouse.com/blog/introduc…
auxten retweeted
Agent memory tip: you probably don't need an ANN index.
Recall is almost always scoped to one user or session. Put that in the ORDER BY key, filter first, and exact cosine over what survives is cheap.
1M memories, 768-dim, on a laptop:
scoped (500 rows) ~40ms
unscoped ~1.5s
auxten retweeted
ClickGraph translates Cypher to ClickHouse SQL, and its embedded mode runs in-process on chDB - graph queries over Parquet/Iceberg/S3, no server. Neat bit: hybrid mode queries remote ClickHouse and keeps the returned subgraph in local chDB.
github.com/genezhang/clickgr…
auxten retweeted
Agent memory is not chat history.
It is traces, recalls, revisions, embeddings, conflicts, and audits.
SQLite is great for checkpoints. Analytical memory needs local OLAP + portable durability.
That is chDB durable: ClickHouse-grade OLAP stays local and embedded; durable state lives in object storage you own.
No DB server. No PVC. No sidecar.
github.com/chdb-io/cookbook/…
Congrats to Hannes and @mraasveldt
Embedded analytics is a category people take seriously now, and DuckDB is the one that opened that road. We met last year and talked for a long time — I still remember that conversation.
Wishing the project a long life in the open, and the two of you room to build what comes next.
Shared at PyCon YVN 2026:
No GIL, Real Gains: Porting a C++ Python Extension to Free-Threaded Python 3.14
youtu.be/soBWS_SlqJc?si=ygv2…
auxten retweeted
This week, we announced that chDB is now available as an ADBC driver.
To install it 👉 dbc install chdb
chDB is an in-process SQL engine powered by the @ClickHouseDB core. Now you can use chDB from any language that supports ADBC.
See the docs and quickstarts below 👇
ALT Diagram showing language clients (C++, C#, Go, Java, JavaScript, Kotlin, Python, R, Ruby, Rust, shell, and TypeScript) connecting through ADBC to chDB, the in-process SQL engine powered by ClickHouse.
Today I gave a talk at PyCon Japan 2026.
The title of the talk was Rediscovering DataFrames: 100× Analytics Without Leaving Pandas
auxten.com/slides/chdb-redis…
