XTDB

XTDB 2.2 will be shipping with an "external source" API: a way to radically simplify how upstream data sources can be plumbed into XTDB. We have developed this API in conjunction with a key design partner on their Virtual Power Plant Orchestration platform, where they have dozens of sources (ingestion apps, Postgres/TimeScale, Kafka topics etc.) and thousands of upstream devices all flowing data into XTDB continuously. This XTDB deployment, running in Azure, is already handling several TBs of data on power readings, device settings, network topologies and more. Why was XTDB chosen? Because the Design Partner needed a radically simpler data model than any alternatives could offer. For them, reliable (correctable!) reporting with total auditability over time series data was non-negotiable. Bitemporal SQL delivers this. #XTDB (more details on the project is linked in the comment below)

The article states that PostgreSQL supports temporal tables via the temporal_tables extension. Frankly, it is not usable for many a reason: it doesn't support Postgres 18, not available in RDS & friends, doesn't use standard syntax, and forces the user to create several triggers for each temporal table. Too many problems, sorry. As the article correctly states, MariaDB has temporal tables, but MariaDB is not MySQL. For those who are interested in temporal tables, I recommend taking a look at XTDB, which was designed to implement bitemporal tables.

"Yes but AI hasn't changed the fundamentals of software engineering". I hear this regularly, and although it's correct it seems to me to be a way of not looking too closely at how much _has_ changed. Christopher Browne built a post-trade equity derivatives risk system in weeks, solo, with Claude Code and Allium specs. The kind of system that normally needs a team and a year inside a bank. I think he's done a great job of capturing this new normal: https://lnkd.in/eKvaV3tt #SoftwareEngineering #AgenticEngineering #FinTech

We're still at the bottom of the S-curve and I'm not sure exactly what's coming next, but many fundamentals are not disappearing soon: AIs need guardrails and good engineering practices matter more than ever. Successfully transitioning a business to survive in the agentic era will entirely hinge on the governance and quality of the connections between the data and the 'token factory', using tools and capabilities that already exist today. The #OSinFinance community is congregating in London on 25 June for #OSFF2026, and I’ll be there speaking about ...AI Readiness! The whole agenda looks pretty great: https://bit.ly/4cdQ68f

We're currently going through some significant under-the-hood changes to XTDB in order to support pulling data in from external data sources like Debezium or Kafka Connect (exciting times! more to come on this soon) but here I want to share a testing technique we're using to unearth potential race conditions in these changes: "deterministic simulation testing" (DST). When you've got a highly-concurrent system, race conditions don't show up every time - in fact, you're often lucky if they show up at all ... well, before production, at least. Race conditions are shy critters - if you attach a debugger, or turn up your logging level, they hide away again. DST changes the maths - architect your concurrent code such that a DST test harness can run it in any possible interleaving. Then, if/when it does find a race condition, the test harness can then give you the random seed of that specific run, so that a test that might otherwise fail intermittently fails every time, at which point you *can* attach debuggers without the error going away. I gave a talk on this at the JUXT Christmas Party: "Chasing Races" https://lnkd.in/extrNZHC

Looking forward to exhibiting with XTDB at #XT26 in the 'Accelerator Room' - also featuring Temporal Technologies. As if the main speaker lineup wasn't amazing enough by itself, we will be showcasing the art of the agentic possible in the corridor track with Structured Product workflows, Customer Onboarding, Risk & Pricing engines, and more! Register your interest at juxt.pro/xt26 - London, June 18th.

Our engineering team has been getting real value out of Claude Code over the past year but the frontier is certainly jagged: - Resolving git conflicts and restructuring branches at (nearly) the speed of thought is a clear productivity boost, unlocking whole new levels of XP workflow - Faster 'typing' and editing is certainly a benefit, but it needs a lot of hand-holding still (because XTDB's code quality has to be protected ...vibe coding is not an option!) - The biggest leverage we're seeing for now is in having the LLM be a super smart rubber duck that quacks back. The conversational and Socratic abilities of LLMs have proven to be a real accelerator in the process of software design (a.k.a. simplification) To this end, James Henderson has published a Claude skill that really helps us with recording and tracking the context behind every PR, without key information getting lost in a sea of transcripts. Context is King.

We're currently going through some significant under-the-hood changes to XTDB in order to support pulling data in from external data sources like Debezium or Kafka Connect (exciting times! more to come on this soon) but here I want to share a testing technique we're using to unearth potential race conditions in these changes: "deterministic simulation testing" (DST). When you've got a highly-concurrent system, race conditions don't show up every time - in fact, you're often lucky if they show up at all ... well, before production, at least. Race conditions are shy critters - if you attach a debugger, or turn up your logging level, they hide away again. DST changes the maths - architect your concurrent code such that a DST test harness can run it in any possible interleaving. Then, if/when it does find a race condition, the test harness can then give you the random seed of that specific run, so that a test that might otherwise fail intermittently fails every time, at which point you *can* attach debuggers without the error going away. I gave a talk on this at the JUXT Christmas Party: "Chasing Races" https://lnkd.in/extrNZHC

Milla Jovovich (yes, the very same one) just released an AI memory system based on "temporal graphs" in SQLite. One of the core queries: `SELECT * FROM triples t JOIN entities e ON t.subject = e.id WHERE (t.valid_from IS NULL OR t.valid_from <= ?) AND (t.valid_to IS NULL OR t.valid_to >= ?)` MemPalace (https://lnkd.in/eAY625zm) follows in the lineage of Zep AI (YC W24)'s Graphiti, which itself puts the bitemporal model front and centre (old facts archived, not deleted). Projects in a similar vein: - MinnusDB / https://lnkd.in/eenFu9Zi - NornicDB / https://lnkd.in/eRgBCBGW - memv / https://lnkd.in/eW8ZDeDF - ctxgraph / https://lnkd.in/ephRzNKX - engram / https://lnkd.in/efV3K7tp - engram (different one!) https://lnkd.in/exx2bFj6 - engram (and another!) / https://lnkd.in/etPsWRDv - Skales / https://lnkd.in/eW396fF8 - agent-recall/ https://lnkd.in/eMyux6-d - Wax / https://lnkd.in/exGSbSYn - Hindsight / https://lnkd.in/e8-upmPF - Daem0nMCP / https://lnkd.in/ec4xsYEr - Memento / https://lnkd.in/eWUxbiHH - CORE / https://lnkd.in/eVtQqwpY - Mem0 / https://lnkd.in/eQAbUenp ...and these are just the ones which have surfaced through my casual monitoring of Reddit over the past couple of months. Much slop amongst those, no doubt. Are there others I should look at? At least some of the buzz seems to have been inspired by this paper: "RAG Meets Temporal Graphs: Time-Sensitive Modeling and Retrieval for Evolving Knowledge" (Oct 2025) https://lnkd.in/eMc8v_sY + video overview by Homayoun S. https://lnkd.in/eCsNFEuf It's fascinating to see all this emerge and it's strong validation for XTDB too, I think. Unlike these other systems though, XTDB is focused (for now) on addressing the bitemporal needs of structured data via SQL, and pushing the costs of bitemporal queries into the index structures as deeply as possible (vs. building on top of an underlying mutable DBMS). Perhaps we'll have to pivot in 2027 if trends continue, or risk having another A-list celebrity eat our lunch 🤔

🗓️ 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.