March 2025 Summaries

Brainstore, a high-performance database built for AI workloads, has become the default in both Braintrust's UI and API. This change is part of an effort to improve performance, with median query times under one second even across terabytes of data. The switch was met with overwhelmingly positive feedback from customers who have opted-in to use Brainstore via a feature flag or API parameter. As a result, the default for BTQL queries against logs and experiments will now return spans by default, rather than traces, although users can still specify otherwise. Self-hosted users are not affected, but will need to enable Brainstore as the default for their users once they're ready. The company is rolling out this change in stages, with some features becoming available earlier than others, and is grateful for customer feedback that has helped develop the best-in-class database for AI workloads.

Brainstore is a purpose-built database designed to handle the high-scale AI workloads that traditional observability tools struggle with. It's optimized for full-text search, has incredible write throughput and low-latency commits, and can scale down to zero or up to hundreds of containers. Brainstore outperforms popular open-source data warehouse solutions and another LLM observability provider in benchmarks, achieving 80x faster performance on real-world benchmarks. Its architecture is designed to take advantage of object storage, with all data living in its own distinct partition and first-class support for semi-structured data. Brainstore has already made a significant impact on several large customers, and the company plans to continue developing and expanding its capabilities, including adding more complex queries, integration with enterprise data lake houses, and hiring additional talent.