How ObsessionDB launched on X
The fastest and most scalable data warehouse for AI
What ObsessionDB launched

Today we are launching @ObsessionDB, the realtime data warehouse for agents. Your agents are hungry for data. Feeding them is getting expensive. Your data warehouse wasn't built (or priced) for that. We learned this the hard way. We scaled our last company to petabytes on https://t.co/WyjakqvQOJ
Transcript
How LaunchVideo.com rated the launch
LaunchVideo.com rating
About the launch
What did ObsessionDB launch? On October 20, 2026, ObsessionDB launched as the real-time data warehouse for agents, as part of Y Combinator's Fall 2026 batch. Cofounder Marc Höffl set out the problem in the launch post: your agents are hungry for data, feeding them is getting expensive, and your data warehouse was not built, or priced, for that. The team says it learned this the hard way while scaling its last company to petabytes. ObsessionDB is a fully managed service built on the open source ClickHouse engine, rebuilt with compute and storage separated. Teams can let ObsessionDB host it on AWS, Google Cloud, or Azure in every public region, or run it in their own cloud account or hardware, with the ObsessionDB team operating it either way. Why do AI agents need a different data warehouse? The founders argue that agents are becoming the heaviest users of the data warehouse. They generate data nonstop, scan it millions of times a day, and even provision their own infrastructure. Warehouses such as Snowflake, BigQuery, and Databricks were built for batch jobs and people running a handful of queries, so they get expensive fast at agent volumes. Operational databases fall short too. The result, the founders say, is that teams end up rationing their data, often dropping raw history to keep the bill down. In the launch video, ObsessionDB claims to be 100 times faster and 20 times cheaper than existing data platforms like Snowflake or Databricks. Those are the company's own figures. How does ObsessionDB work? The core idea is simple: store data once in object storage, then run stateless compute nodes that read from it. Storage grows by adding bytes, and compute grows by adding nodes in seconds rather than days. A cache mesh keeps hot data on local disk, so queries do not wait on a round trip to object storage. ObsessionDB keeps ClickHouse SQL, functions, and drivers, so it works as a drop-in replacement. The architecture also isolates workloads. Separate compute pools for ingest, real-time APIs, and analytics mean a heavy warehouse query cannot slow down a production API. For agents, ObsessionDB adds isolated read-only compute over the same data, full logging of every query an agent runs, open agent skills through a toolkit called CHKit, MCP access from tools such as Claude, Cursor, and Codex, and, in early access, a sandbox branch of production data for each agent run. Pricing is based on compute hours, not on the number of questions asked. Where did the idea for ObsessionDB come from? The founders previously built Numia, which runs blockchain data infrastructure at near-petabyte scale. In a January 2026 post, Höffl explained that Numia ran ClickHouse for real-time APIs and BigQuery for warehouse work, with 47 synchronization jobs between them. Adding a ClickHouse replica meant copying 25 terabytes and waiting three to four days. After rebuilding the system with separated storage, Höffl says monthly infrastructure cost fell from 8,000 dollars to 2,000, adding capacity went from days to seconds, and the 47 sync jobs dropped to zero. He was also candid about the tradeoff: queries not in cache add about 5 to 15 milliseconds, and the benefits show up at scale rather than on small datasets. Who uses ObsessionDB? ObsessionDB says it is processing millions of real-time API requests for customers, turning AI agent session transcripts into live, queryable data. Its biggest customer moved off Elasticsearch and serves real-time APIs on 50 terabytes of data, ingesting hundreds of thousands of events per minute. On its site, Range says it screens more than 90 blockchains in real time with sub-400 millisecond p99 latency, and Binary Builders says it onboards customers at 40 percent lower cost than its previous ClickHouse setup. A free developer tier is available with no credit card required. Who founded ObsessionDB? Rafael Aviles, Marc Höffl, and Marc Leuthardt are the cofounders. Höffl was previously CTO at Numia, and Leuthardt cofounded Numia. How did the ObsessionDB launch perform on X? Höffl posted the launch video from his personal account on October 20, 2026. The post reached about 27K views, with 307 likes, 35 reposts, 59 comments, and 179 bookmarks, as recorded on this LaunchVideo.com directory entry. He had only about 330 followers at the time, so the post reached roughly 80 times his own audience. The film runs about 58 seconds and opens with a question: if your agents are hungry for data, can your data warehouse feed that appetite? LaunchVideo.com rated the launch 3.0 out of 5, with 4 out of 5 for editing and copy, 3 for hook and sentiment, and 2 for distribution. What can founders learn from the ObsessionDB launch video? The ObsessionDB launch video shows how a founder-led launch video can lean on lived experience. Saying the team scaled its last company to petabytes and watched the infra bill explode gives the product instant credibility with data engineers. The bold 100 times faster and 20 times cheaper claim is the kind of line that gets attention, but it lands harder with proof on screen, such as a cost chart or a real query running. ObsessionDB has that proof in its own before and after numbers from Numia. For founders selling infrastructure, the lesson is to pair a big claim with a real story and real numbers. LaunchVideo.com helps founders build launches around that kind of evidence.
Tags
Questions
Frequently asked questions
The ObsessionDB launch post reached 27K views on X, with 307 likes, 35 reposts, 59 comments, 179 bookmarks.
ObsessionDB launched on X on October 20, 2026 with the message "The fastest and most scalable data warehouse for AI".
What did ObsessionDB launch?
LaunchVideo.com rated the ObsessionDB launch 3.20 out of 5: hook 3/5, editing / creativity 4/5, copy 4/5, sentiment of launch 3/5, distribution strategy 2/5.







