Artie

Artie supports Apache Iceberg as a destination.

Iceberg stores your data as open tables on object storage. Keeping those tables in sync with a production database means handling:
> updates and deletes
> schema evolution
> catalog commits and snapshot conflicts
> staging and compaction

Artie handles the write path. Replicate production databases into Iceberg tables through S3 Tables or an Iceberg REST Catalog, then query them through the engines your catalog supports.

Keep the flexibility of Iceberg without building and operating the CDC-to-lakehouse pipeline yourself.

Thank you to everyone who came out to AI Demo Night on Tuesday for SF Tech Week.

8 developers and founders demoed what they've built, and most of the night was about agents: how people are building them, running them, and putting them to work.

Our founding engineer @anirudhsriramzz showed how we run Hermes, an open-source agent harness, at Artie with separate agents for sales, engineering, and design.

Thanks to all the speakers, and to @ClickHouseDB and @llama_index for co-hosting with us.

Firecrawl is the API for the web. More than 1.5 million developers build on it.

The company is growing fast, and the data is growing with it. With one data engineer, Rahul Singh, @firecrawl uses Artie to move around 4 billion rows a month from Postgres into BigQuery. That data feeds every dashboard, usage report, and spending decision the company makes.

Since Firecrawl started on Artie in December 2025, it has raised a $75M Series B, and the pipeline hasn't had to change. When Rahul moved a large share of the data from Supabase to PlanetScale, he used Artie for that too, and the replication side was done within two days.

In his words: "It just runs in the background, and I don't have to worry about it."

SF Tech Week starts today 👀

Tomorrow, Artie is co-hosting an AI Demo Night with @ClickHouseDB and @llama_index at CANOPY Jackson Square.

Come watch 8 developers and founders demo what they've built across the AI stack. Our founding engineer, @anirudhsriramzz, will be going first.

RSVPs are full, but you can still join the waitlist. Link below 👇️

Aurora PostgreSQL can now query Apache Iceberg tables in S3 directly. You can join live Postgres rows with years of history in your lake in a single query, without a pipeline copying that data back into the database. It uses DuckDB embedded in Aurora, following AWS’s acquisition of DuckLabs (the team that maintains DuckDB).

This solves the read side. But something still has to keep those Iceberg tables current, through schema changes, backfills, and high write volume. That’s what we do at @artie_labs – we stream changes from operational databases into Iceberg-backed tables in Amazon S3 Tables, so the history Aurora reads stays fresh.

momogood's Redshift warehouse used to fall up to 24 hours behind during fundraising surges. With Artie, it stays within minutes, even at 10M rows an hour.

momoGood, formerly Tatango, is an AI-native giving platform for nonprofits. More than 3,200 organizations use the platform, which has helped raise $2.4B.

For momoGood Messaging, fresh data is operational. Sending activity feeds donor retargeting, campaign reporting, customer-facing metrics, and billing. And volume isn't steady: fundraising deadlines and urgent events can push 10M rows through in an hour.

Before Artie, the team replicated Postgres and MySQL into Redshift with AWS DMS, tuning each pipeline by hand. Under peak load, DMS fell behind or stopped entirely, leaving retargeting, reporting, and billing working off stale data.

Now Artie replicates their critical tables to Redshift through CDC:
> Minute-level latency, including on peak sending days
> 600-800M rows in a peak month, no manual backfills
> Lag and root cause visible, so fixes can be verified
> ~80 dev hours a month of data firefighting gone

Full case study below👇️

We're hosting an AI Demo Night with @ClickHouseDB and @llama_index next Tuesday, October 6th, during SF @Techweek_.

8-10 developers and founders will show what they built across the AI stack. There will be dinner and drinks after.

📍CANOPY Jackson Square

Artie supports @supabase as a source.

Turn the Postgres database powering your Supabase app into continuously updated data in your warehouse, lakehouse, or another operational database. Artie captures inserts, updates, deletes, and truncates through Postgres logical replication.

Your team gets production-ready CDC without building export jobs, polling APIs, or operating a Kafka/Debezium stack.

Artie is at Rows & Columns Summit today at the Contemporary Jewish Museum in San Francisco!

At 2 PM, our founding engineer William Haw will join James Sutton, CTO of ShipScience, for a talk on how ShipScience moved 3 billion MySQL row updates per month out of production and into MotherDuck with Artie.

If you’re at the summit, come say hi!

Artie will be at Rows & Columns Summit tomorrow at the Contemporary Jewish Museum in San Francisco!

If you're running transactional and analytical systems side by side, this is where the operational lessons show up: where the OLTP/OLAP boundary actually falls, what breaks when data moves across it, and how teams are running these architectures in production today.

Speakers include:
> @andy_pavlo from @ClickHouseDB
> Nikita Shamgunov from @databricks
> @kiwicopple, CEO & Co-Founder of @supabase
> Hannes Mühleisen, Co-Creator of @duckdb
> William Haw, Founding Engineer at @artie_labs

Standing up a production CDC pipeline used to mean designing source access, replication, backfills, destination writes, schema-change handling, monitoring, and recovery.

Now, it can done with a single prompt: “Create a pipeline from the staging Postgres database to Snowflake.”

With Artie MCP available in Claude Code, Cursor, and Codex, an AI agent can use Artie’s tools to configure and manage that workflow. So engineers can start with what they need, instead of spending months assembling and maintaining the pipeline themselves.

Artie 🤝 Hermes

Artie now exports pipeline metrics to @Grafana Cloud Metrics.

Teams already using Grafana Cloud can monitor database, Kafka, warehouse, application, and now Artie pipeline signals in one place. Artie exports six metrics directly to Grafana Cloud’s Prometheus remote-write endpoint, without needing to deploy Grafana Alloy, Prometheus, or a separate collector yourself:

☑️ Rows processed: pipeline throughput
☑️ Ingestion lag and row lag: whether data is arriving and becoming available on time
☑️ Flush time and flush count: write behavior
☑️ Replication slot size: PostgreSQL replication health

Compare Artie ingestion lag with source-database and warehouse signals when you need to investigate a data freshness issue. Or create a Grafana alert when a lag, flush, or replication-slot metric crosses a threshold your team defines.

Import the Artie Overview dashboard to get started, then adapt it to the Grafana workflow your team already uses.

We'll be at Rows & Columns Summit on September 22 at the Contemporary Jewish Museum in San Francisco.

If you're running transactional and analytical systems side by side, this is where the operational lessons show up: where the OLTP/OLAP boundary actually falls, what breaks when data moves across it, and how teams are running these architectures in production today.

James Sutton (CTO, ShipScience) and William Haw (Founding Engineer, Artie) will be speaking on what's actually become possible with AI and data this year.

The summit is free to attend – register today!

@Substack was Artie’s first customer.

At the time, we had never operated at their scale. There were real hiccups early on. Mike Cohen, Substack’s Head of AI & ML Engineering, was direct about what needed to work. We took the feedback, shipped fixes quickly, and a lot of Artie’s reliability today traces back to those early weeks.

Today, Substack replicates roughly 5 billion rows per month through Artie.

We’ve renewed our partnership with @Snowflake as an AI Data Cloud Products partner.

More teams are pushing Snowflake past reporting and into production AI, building agentic workflows, fraud detection models, customer personalization, and more. But these workloads break the moment the data underneath goes stale.

Artie streams production data into Snowflake in real time, keeping the warehouse continuously current for decisions and workflows – all without infrastructure to manage.

Teams can purchase Artie with Snowflake MCD, so procurement runs through an existing Snowflake relationship instead of a separate contract.

Postgres 17 changed an important part of the CDC failover story: logical replication slots can now survive a primary promotion.

With HA configured, a logical slot can sync to a standby and survive promotion, preserving Postgres’s record of where CDC left off. But it does not preserve the consumer's TCP connection or the consumer's offset state.

Artie now adds the other half: Artie Reader persists its progress and recovers automatically when the writer endpoint moves.

Postgres preserves the slot, while Artie preserves the consumer’s progress. Together, they make CDC recovery more resilient during database failovers.

Full breakdown below 👇️

Artie supports @ClickHouseDB as a destination.

Stream real-time changes from Postgres, MySQL, MongoDB, and other operational databases into ClickHouse without building a custom ingestion pipeline.

Artie maps CDC changes into ClickHouse’s ReplacingMergeTree model, including version and delete markers, so analytical tables can reflect source updates and deletes without a separate merge process.

Initial load, continuous change capture, and destination schema handling are all handled by Artie – no infrastructure for your team to run or maintain.

ICYMI: Artie now has a free plan 🎉

Startups are operating workflows around production data with smaller teams than ever. If they want that data in their warehouse (or another database) in real time, they shouldn’t have to stand up and operate Kafka, Debezium, connectors, and monitoring just to test a real workload.

We launched Artie’s free plan so you can get real-time data replication without a credit card, sales call, or multi-quarter CDC infrastructure project. Free for up to 2 pipelines and 1 million rows a month. Scale when you need to.

Try it: go.artie.com/free

We benchmarked Artie and AWS DMS on real-time CDC from Postgres to Snowflake under sustained writes of ~22,400 events/sec for 10 minutes.

Using history mode, which writes to an Slowly Changing Dimension (SCD) Type 2 table, DMS had 8.5× Artie’s latency nine minutes into the run. Its latency reached 1,980 seconds by the final reading, about 33 minutes behind the source. Artie’s final reading was 97 seconds.

At sustained load, Artie’s latency stayed bounded while DMS continued to fall behind.

Full breakdown in the comments.