SpacetimeDB

Introducing SpacetimeDB 2.0. Web development at the speed of light. Come learn what fast really means.

@vmf____ @JasonBud @XFreeze Amen!

@JasonBud @XFreeze @SpacetimeDB

Grok expansion this month is unbelievable Most people still do not realize what is happening Grok is being plugged directly into the tools people already use every day: • Interactive Brokers — portfolio analysis, market research, strategy + order instructions • AWS Bedrock — enterprise access to Grok 4.3 • Databricks — AI agents + data workflows • Microsoft Word, Excel & PowerPoint — office productivity • Vapi — voice agents • eToro — real-time market sentiment inside Tori • Gopuff — AI shopping assistant • Warp — coding + terminal workflows • T3code — AI coding-agent workflows • Vercel — deployments, build status + domains • MongoDB — database management + queries • Sentry — debugging + error analysis • Cloudflare — Workers + infrastructure tools • Chrome DevTools — browser control + performance testing • Firecrawl — web search, scraping + page interaction • Superpowers — agent workflow plugins And that is not even the full list Grok Build also added /goal for long-running autonomous tasks, Agent Dashboard for managing multiple coding sessions along with Grok Build 0.1 & Grok Composer 2.5 Grok Imagine Video 1.5 is now faster, better, and live across API, web, iOS, and Android Grok Connectors already plug into Google Workspace, Outlook, SharePoint, OneDrive, Notion, GitHub, Linear, and custom MCP servers Finance. Cloud. Coding. Voice. Shopping. Data. Office work. Browsers. Developer tools. Agents Grok is moving from “AI you chat with” to “AI that works inside everything” xAI’s execution speed right now is insane

For some Friday afternoon fun, allow me to hijack your productivity with this game: Escape the Moon! It started as a side project to kick the tyres on #SpacetimeDB, but snowballed into a whole cards-and-crafting web game. escapethemoon.vercel.app

@Bradley_Moore @jhleath How about isync?

@jhleath fsync

i finally got nerdsniped by the spacetimedb stuff, so i spent an hour with claude last night to optimize the postgres test and was able to do 550,000 operations/s on a PlanetScale metal -- and we could probably do more if i had time how? now, remember, i don't understand SQL databases [just storage] so take all of this with a grain of salt, it won't be up to the benchmarking standards that @TylerFCloutier @samlambert or @BenjDicken would hold themselves to so the first thing that i did was dig into the numbers. the spacetime benchmark promises 300,000 operations/s over an iodepth (number of outstanding requests) of 2,560 (64 clients, 40 outstanding requests each) if you math that out, in order to achieve those numbers, every "io strand" only had to complete 100 operations/s (~10ms latency budget per operation), which [to me] seems well within the realm of possibility for any database so i started testing it out. at first, i tried the benchmark as-is and [of course] got around 400 op/s on postgres (on my laptop or planetscale), this said to me that something else was going on. first, i asked claude whether or not the SQL statements were what you would write if you were optimizing for performance, and the answer (of course) was no -- rather than selecting the data to the client, then updating it in separate statements, we could collapse that into a single statement. i tried a large planetscale instance (M-256 [$10K/mo, i know] instead of M-10) and this allowed the throughput to jump from 400 op/s to 5K op/s. cool, but not cool enough. the question remained, though, the fundamentals here aren't crazy -- something must account for the difference in IOPS. since the spacetimedb code runs on the server-side, i figured that apples-to-apples would be to have a server between the client and the postgres database too. this would allow us to coalesce and batch transactions just like spacetime db (with no client acknowledgement until things were durably committed) this was a big jump. we went from 5K op/s to 113K op/s, while keeping the actual load on the database itself super-small. this makes sense, i think the whole point of spacetimedb is that you can do a lot more per-unit-work like this. there were a few more minor optimizations (moving to an append-only log instead of calling UPDATE on every changed row, and then committing that log to postgres in parallel), but it was otherwise straightforward to get to 500K+. so, what did we learn? i think that everyone in this argument is right to some extent. @PlanetScale is correct that benchmarking Postgres the way that the Spacetime people did wasn't fair to the capabilites, and @spacetimedb is correct that their architecture provides a meaningful throughput improvement (and that multi-tenanting these workloads together will be lower cost for people who need to burst) it helped me to contextualize that Spacetime isn't actually a *database* (similar to Convex), it's really a *framework* over a database which allows certain classes of workloads to get higher throughput because they're easy to coalesce. i understand why Spacetime doesn't necessarily market themselves (or even sell the product this way), but it says to me that there's probably a market opportunity for the "1,000x Postgres performance" framework that just handles all the wasm compiling and running for you all in all, a fun way to spend an hour with claude

Benchmarking @spacetimedb is just so good. Our TypeScript number is pretty wild (303,920 ± 4,712 transactions/sec), even faster than Rust. The main crux is that we improved our V8 threading model a lot. The TypeScript modules are running inside SpacetimeDB on V8, so don’t think of this as a Node.js app talking to a database. Instead of making every reducer call pay the cost of submit → wait → wake async task → hand off and submit the next call, we now push reducer work into a hot FIFO V8 lane where we have a dedicated worker thread owns the V8 isolate and keeps draining that queue. So V8 stays warm, reducer code stays hot, and the next transaction can start right after the previous one finishes without waiting for some outer task to hand the instance back. At this scale, small overhead is not small. When you are doing ~300k transactions/sec, a few microseconds of scheduler wakeups, thread handoffs, or synchronization can decide the whole benchmark. At first I thought since when typescript got faster than rust. Haha. If you keep app logic close to the database, remove the normal app-server/database round trip, and make V8 behave like a continuously running transaction engine, TypeScript can actually fly high like an eagle.

@59thProfile There isn't even anything here to ignore

Ignore this @spacetimedb

lol. Lmfao even. The entire software industry is a joke. Just wow

@richardsolomou @posthog Node sidecar works well, but you should try procedures too!

liked it enough that i shipped an official posthog example for spacetimedb the same week 😄 js + node, all keyed on the spacetimedb identity so events stitch to one person. the fun puzzle was backend capture - reducers are sandboxed so no network, so i did it with a node sidecar that subscribes to the db. live for your users now: github.com/PostHog/postho… happy to walk your team through it if it's useful 🙌

set out to build an MMO. 6.5 hrs later i have a field of hedgehogs wandering around a cave and honestly i'm not mad about it. this week @posthog code rewrote the backend to @spacetimedb, then added pixel terrain, grid movement, pushable boulders, discord login + the hogs. what is the game now? unclear. it's great though. stream 2 of many. round 3 thursday 🦔 🔴 vod: youtube.com/watch?v=hQO-yd…

Using @threejs and @spacetimedb to create my MMORPG, Dungeonpunk LITE, has been a delight. A new content authoring tool: Mob Encounter Zones. Placing, shaping, picking a mob, setting pop-limit, export changes and bang, it's in game.

You can integrate any custom network solution you want inside @SomniumSpace so I decided to add @spacetimedb inside my MMORPG world "Blades Edge" with the helpful hand of @BentlyBro_AGPT (legend btw) Things got very exciting, very quickly 😂

Clockwork Labs has been a remote company since day one. I'm quite pleased that our employees are trustworthy adults who are competent enough to manage their time appropriately.

@tsurg_ Ah, but it is consistent. We could change the capitalization

@spacetimedb Honestly, It looks worse. reads like "space timed b"

Damn it feels good to drop that underscore.

@thekitze Tell Elon Musk, they stole ours

@spacetimedb now to stop using the grok logo and all good

@steipete @geldata You're missing @spacetimedb!

Anyone in my timeline who did more than playing with @convex or @geldata ? These seem like the future “ORMs”. (We need a new word for that, also am I missing a contender?)

Check out @DevanshuXi's dive into SpacetimeDB!

The fact that there is no distributed operating system is a tragedy for mankind. And it’s because Unix’s “everything is a file” abstraction was just good enough to survive networked systems. Unix is too good for its own good.