My talk from @sync_conf 2025 is ready!
Check it out to learn how:
1) CRDTs + cryptographic permissions work
2) @jazz.tools is now a general-purpose database
3) our unique Jazz Cloud infra compares to
- traditional stacks
- other sync engines
- Durable Objects
www.youtube.com/watch?v=wils...
it has begun
Btw: this is one of the first third-party libraries for Jazz ever.
And if it's usefulness, API design, documentation, demo and attention to detail are any indication, we're going to have an *insane* ecosystem.
Stay tuned 🎶
And it elegantly inherits all the local-first properties! 🤯
- vector search incurs no network latency at all
- vector search is offline first
- real-time cross device sync, with new matching items appearing in search results reactively as they are created (watch the video)
People keep asking us if @jazz.tools will have vector search...
What's better than building a first class vector search feature for Jazz in maybe a couple months?
A single person right now being able to ship it as a small, very neat, very high quality library *on top of* Jazz!
vimeo.com/1111603758
LLMs are a crystallisation not only of human knowledge but also of the human condition
I think there is immense value in them encoding both
- the surprising commonalities in our experiences, struggles and hopes
- *and* the myriad dimensions of individuality
Proud for the event to be co-sponsored by Vercel!
You'll also hear from Luba Kravchenko about Cache, CDN and Fluid Compute which give you much more flexible, cost-effective and performant versions of traditional infra abstractions.
One thing that's clear is: there is actually a lot of pressure on app builders because app requirements have become so much more complex.
We think it's time to rethink the stack.
Given our realisation that *actually, @jazz_tools is a database* (see QT)...
...we're hosting another meetup in SF this Wednesday!
(RSVP below!)
Thinking of Jazz as a DB begs the question: how does it fit into the stack?
bsky.app/profile/jaz...
narrative update
Jazz isn’t replacing databases and backends
Jazz isn’t local-first only
Jazz isn’t a sync engine only
Jazz is simply a *new kind* of database
🧵
FUTURE OF CODING LONDON
is next week!
Last chance to submit your awesome demos!
lu.ma/lo2medup
Relatable
This is why I think The GP’s Speed Trials are the most interesting codegen hackathon yet. A great chance to demo and share what works in practice, and help shape the alchemy that is quality engineering at breakneck speeds. Proud to be a tooling partner as @jazz.tools
b) using their ability to *review* code deeply for correctness and test coverage
I’ve been using “vibe coding” in a looser sense than Karpathy’s “fully blind” definition, for good reason.
The biggest impact of AI tooling that I’ve seen is senior engineers
a) using their ability to clearly specify what they want and *build* extremely quickly with LLMs
The wait is finally over!
It’s time for another
FUTURE OF CODING LONDON
Tuesday, July 22 @ 18:00
Kindly hosted by Granola at their office
Sponsored by @jazz.tools
RSVP here:
lu.ma/lo2medup
Demo/Talk submission here:
foc-london.notion.site/21c379f0e4c...
Lots more interesting thoughts on this soon, but as I get deeper into fully auto-running agents, a simple realisation (and I cannot overstate this):
Vibe coding is the first time
computers have felt as magic again
as they have
when they first felt magic to me
Two weeks after alpha-3 which foreshadowed our move to @jazz.tools for greater stability, alpha-4 gets it done!
blog.muni.town/roomy-space-...
Special shoutout to @flo-bit.dev who took the lead on this release.
Up next is bridging so Roomy & Discord (later Matrix) folk can share the same space.
We’ve got a place and I think a date in ~1 month, will DM you @todepond.com, stay tuned for the announcement @amb26ponder.bsky.social
Watch @anselm.io talk about how local-first is accidentally perfect for the AI age. youtu.be/e3-yIWGNBLQ?...
My full talk from @localfirstconf.com is up!
Featuring 1y of progress on @jazz.tools
And: two big announcements + the most sketchy, yet successful demo I've ever done.
- Waffle, a Discord/Slack alternative built on Jazz
- Soul, our upcoming AI toolkit for Jazz
www.youtube.com/watch?v=e3-y...
> The AI just is "one more collaborator" and you just dump the tokens into Jazz state as they come in.
This is the correct mental model
Resumable streams? (Something that other solutions need lots of extra infra for, see OP)
Obviously built-in. The same sync protocol that gracefully deals with offline or sketchy connections (for any kind of Jazz app feature) handles network interruptions during the LLM response.
Jazz takes care of syncing the tokens to everyone who's interested in them, globally, at super low latency.
They're also immediately persisted as a typed CoValue (typically co.richText) with rich edit metadata and permissions - ready to build more complex multi-user apps around.
So everything in Jazz is architected around that requirement, from the client to the distributed sync & storage infrastructure.
This is a godsend when you start doing AI stuff. The AI just is "one more collaborator" and you just dump the tokens into Jazz state as they come in.
Streamed LLM responses are the perfect example of why we didn't base @jazz.tools on an existing DB, but are building a new backend + DB from scratch.
Originally we just wanted to support Figma/Notion-like apps where we knew we had to sync 100,000s of tiny edits per second... 🧵
x.com/marbemac/st...
To be fair, this makes Jazz's scope much bigger. There are still areas we need to catch up on, like queries/indexing over lots of data.
For a lot of Jazz apps it's surprisingly feasible to do brute-force filtering, but we are going for full feature parity with traditional DBs!
Finally, and most importantly @jazz.tools will prove that the local-first ideals are not "lofty goals for a niche of apps" but pragmatically achievable and obviously desirable for *every* app.
And it's already becoming clear that going all-in results in better UX *and* DX!
