XTDB

🗓️ XT26 in London on June 18th 🗓️ It's got XT in the name so naturally we'll be there to chat databases and time travel! To be clear, this is not "the XTDB conference" (XT16 conference pre-dates the database) but the lineup is looking great.

There is a temptation, when describing a system like XTDB, to reach for grand metaphors. But the value is quieter than that. It is the simplicity of asking temporal questions in SQL rather than building temporal logic in application code. The particular calm that comes from an audit trail that cannot be edited, because the database will not permit it. And when your data arrives from five different systems, each with its own notion of when things happened, it is the confidence of knowing those timelines are preserved, not flattened into the moment they arrived. Other systems let you query the present. XTDB lets you query the truth, which includes the present, but is not limited to it.

What makes a *good* decision? The word 'decision' originates from the Latin 'decisio' ("a cutting off, separation, or settlement"), reflecting a metaphorical action of cutting away other options to reach a conclusion. To properly audit the quality of a decision requires not only verifying every step of the processes surrounding it, but also careful counter-factual analysis of the options which were cut away. As we delegate more decision making to AI agents, how can software infrastructure empower organisations to easily assess whether the decisions being made are any good? And how do we establish that our AIs are improving over time by learning how to make better decisions? Decision auditability is the central capability that XTDB provides, and it takes the form of *bitemporal SQL* - a declarative language for retrospectively understanding the decisions made across your organisation with microsecond clarity. A strong foundation for continuous learning, with or without AIs.

Huge congratulations FoundryOS team for reaching your go-live milestone! We're thrilled that XTDB is a core pillar of your release here, helping to solve the pains of complex KYC audit requirements. (...new XTDB case study coming soon! ✨)

To audit is to verify and scrutinise decisions. The word 'audit' originates from the Latin term 'audītus', meaning "a hearing" or "a listening". In early accounting, audits were not just written, but involved accountants reading their records aloud to an auditor to verify accuracy. As the world's information grows rapidly more complex (and, simultaneously, more full of errors...), how can our software infrastructure best support the imperative to verify the systems and processes that really matter? XTDB's answer to this essential problem of 'decision auditability' is *bitemporal SQL* - a declarative language for retrospectively understanding the decisions made across your organisation with microsecond clarity. To hear the signals amidst the noise.

Will AI-assisted engineering spell the end of users hearing "sorry, I can't restore your data" from support teams? Or will IT organisations now have to work extra hard to prevent this gap from widening as the amount of code shipped accelerates? One sure bet is that engineering teams who can lean on their database infrastructure to transparently preserve all history will avoid burning untold tokens on the soft-delete problem, over and over again. Regulators and compliance teams would be much happier to see that happen also!

Our second biggest release of the year just dropped - XTDB 2.1.0 - featuring some pretty cool bells and whistles. At the top of the list is 'multi-database' support where you can ATTACH databases on the fly and query across them. I just wrote a post on it: https://lnkd.in/eSJnxsce Unlike Postgres or other SQL databases which tightly couple storage, compute and replication, XTDB makes it (almost) trivial to scale out read & write access to secondary databases across an organization. Think Data Products in some kind of Data Mesh. Normal database silo bottlenecks, where some privileged nodes own the storage layer (and everyone gets a vaguely consistent read replica, if you're lucky), just don't apply in our shared storage architectural model. From this perspective, XTDB is closer in spirit to the Open Table Formats (Hudi, Icerberg, Delta) than the OLTP monoliths.