Matt Galligan

40.9K posts

Matt Galligan banner
Matt Galligan

Matt Galligan

@mg

Dad to four kiddos, Builder, @XMTP_ co-founder

เข้าร่วม Temmuz 2006
1.6K กำลังติดตาม31.1K ผู้ติดตาม
ทวีตที่ปักหมุด
Matt Galligan
Matt Galligan@mg·
I’m not a Rust dev. Hell, I’m not an engineer at all. But I'm excited to ship BLZ—a CLI for caching and searching llms.txt docs in ~5-15ms on average. Think ripgrep, purpose-built for llms.txt docs—offline, deterministic, real fast.
Matt Galligan tweet media
English
4
3
29
3.8K
Matt Galligan
Matt Galligan@mg·
Been fun to try to get it all figured out. Filed a few fix PRs along the way. I could certainly be wrong here but it seems like maybe the SDK is a fork of the official (hosted) Proof. You might consider flipping that, so that the hosted version consumes the SDK. This would mean the community could add fixes and features, and then you could pull them in to hosted. Of course, the hosted version could have things to contribute back to the SDK, or you could keep some things separate. Setting it up in this way though means you could benefit from community members fixing things up for you too.
English
1
0
1
19
Matt Galligan
Matt Galligan@mg·
Zellij is great. Have been running on it full time for about 6mo since leaving tmux behind. Only thing to be mindful of is if you've got a session running, with a bunch of tabs/panes, and fire it up on desktop and mobile, you might run into some crashing…and then all of your sessions stop cold. Have had it happen more than enough times and have had to be more delicate than w/ tmux as a result. But the UX is far better.
English
1
0
1
2.5K
Matt Galligan
Matt Galligan@mg·
Been working on my "MG Voice" agent skill, which guides an agent to write more like me. Added in some linting rules that are deterministic vs. just relying on vibes. So naturally Claude is testing them. And…well…hilarity ensued.
Matt Galligan tweet media
English
0
0
5
375
Matt Galligan
Matt Galligan@mg·
@danshipper @kieranklaassen @every @CloudflareDev Looks like the Proof SDK doesn't yet have a deployment guide, so will be adding that to the draft. Including both Express/Railway and Cloudflare deployment instructions, so it should be straightforward to get going w/ it.
English
1
0
0
37
Matt Galligan
Matt Galligan@mg·
@ejae_dev Thats by design! Two different problems. DOs own the doc-level hot path (collab, marks, sync). D1 sits above as the shared catalog for cross-doc queries. DOs fan out metadata on mutation; D1 handles search, analytics, workspace views. Per-doc isolation is the feature not the gap.
English
0
0
0
61
ejae dev
ejae dev@ejae_dev·
@mg durable objects are perfect until someone needs cross-document search or global analytics. each doc being its own island is great for isolation but painful for anything that spans the whole workspace.
English
1
0
0
82
Matt Galligan
Matt Galligan@mg·
@danshipper @kieranklaassen @every @CloudflareDev Not sure! Everything was wired up to be a 1:1 parity with the Express server setup, which did have RBAC and such, but no explicit login functionality exposed in the SDK. But Workers can handle auth no problem, bunch of ways to add it.
English
1
0
1
45
Matt Galligan
Matt Galligan@mg·
Haven't been able to test that much in the wild yet but most fully cold starts should be ~10-30ms. Docs reconnecting with lots of pending changes (>200) might be ~50-100ms? Regardless, it should be really fast.
English
1
0
1
227
Matt Galligan
Matt Galligan@mg·
@trq212 Project-level agent memories could be great if they lived on a separate worktree, and regularly synced. As is, there are some challenges with versioning them. But teams should be able to benefit from them! Beads had this pattern to sync its db and it was really clever.
English
0
0
0
158
Matt Galligan
Matt Galligan@mg·
@joshu Would love to see that too, but haven't found anything yet. For now I've printed these, which have worked quite well, and kept things dust free. But they're too big to fit inside of a wall box of any sort. printables.com/model/845688-w…
English
0
0
0
178
joshua schachter
joshua schachter@joshu·
is there such a thing as multi butt-splice wagos? i want this but for 4-8 SEPARATE signals (all in one physical connector)
joshua schachter tweet media
English
8
0
8
2K
Matt Galligan
Matt Galligan@mg·
@kepano have a bug report for you re: the @obsdmd CLI tool. Tried installing on two other machines of mine today, and their worked. Kept saying "Unable to connect…" It took a few agents hammering on ideas, but finally cracked it thankfully: gist.github.com/galligan/fd619…
English
1
0
0
329
Matt Galligan
Matt Galligan@mg·
Alright, new feature is getting built…by a non-deterministic agent. Agent builds its own tests first then codes the feature until green. So…its tests pass. CI says tests pass. But was it the right solution? Did the skill that's sitting in the agent plugin which teaches how to use your tool also get the critical update with the new feature? Some things are just squishier. And having a code review agent as a backstop has been invaluable for me.
English
0
0
1
34
Zain Hoda
Zain Hoda@zain_hoda·
@mg Tell me more? I feel like this should just be part of deterministic tooling.
English
1
0
0
36
Zain Hoda
Zain Hoda@zain_hoda·
I don’t understand why people do AI code review when you can just ask your coding agent to set up GitHub workflows to run tests. Is there some additional benefit I’m not seeing?
Zain Hoda tweet media
English
2
0
3
325
Matt Galligan
Matt Galligan@mg·
I've *really* liked using @greptile and it's been the best code review agent I've used yet. However, this pricing change is a doozy. In my workflows, I put up smaller PRs in stacks, but that means more of them. $1/review past 50/mo will be blown past in less than 3 days. In the last 30 days, my agents have put up 571 PRs for just a subset of my repos. Under the new pricing structure what was once $30/mo now becomes on more on average than what I pay for my Claude Max + ChatGPT Pro plans *combined*. Pretty bummed out by this change tbh. I'm sure I'm in the 10% of users @dakshgup describes here, so $30/mo probably doesn't even cover cost. But going from $30/mo to more than $500/mo is not gonna work for me. 😵‍💫
Daksh Gupta@dakshgup

Lastly, we’re modifying our pricing to a base + usage model, similar to many popular AI coding tools. The new pricing is $30/developer/month, which includes 50 reviews, after which reviews cost $1 each. Less than 10% of active users will exceed the included usage. Existing users will remain on the existing pricing until the next billing month. Better coding agents have led to a drastic increase in the number of commits reviewed by Greptile among power users. This new pricing model allows us to focus solely on providing an uncompromising AI code review experience.

English
0
0
10
2.6K

ค้นพบ

@kieranklaassen @every @CloudflareDev @danshipper @ejae_dev @trq212 @joshu @elonmusk