cvberens

Building Avrea. Faster CI, shorter feedback loops, better infrastructure for humans + agents.

[cvberens @iamBrns]

@roncohen @bentlegen agree, and culling is hard because nobody owns those tests. the cheap first step is caching and only running affected tests, then prune what never fails.

[cvberens @iamBrns]

@prvnsmpth @deedydas agree, fix the pipeline first. but agents push to CI just to find out if their code works, and there's way more code now. Anthropic said their CI jobs went 25x in 6 months. then GitHub's own runners are the priciest way to run it. Avrea and others are cheaper per minute.

[cvberens @iamBrns]

@deedydas @RunsOnRunsOn @starslingdev @WarpBuilds @UbicloudHQ @trylle @incredibuild @rwx https://t.co/UYbNSHzVRR?

[cvberens @iamBrns]

@leeschmidt123 @Altimor 20 → 2 min is great. what made the biggest difference?

[cvberens @iamBrns]

@leiferik @Altimor what's the slowest part for you, queue, setup or the tests themselves? usually one of them is most of it.

[cvberens @iamBrns]

@YueSh5 @Altimor 30 min per change usually means everything reruns from scratch. quick wins: cache deps and build outputs, shard tests across parallel jobs, only affected tests on PRs and full suite on merge. Do a quick benchmark om https://t.co/CqNRBYYOI5

[cvberens @iamBrns]

@GaleGiddy98356 @Altimor if you self-host, what bites later is ops: clean ephemeral envs, cache invalidation, queueing when every box is busy. plan for those upfront. (I'm at Avrea, we run managed runners, happy to share what we've learned)

[cvberens @iamBrns]

Norma is a personal assistant you just text. I'm on the list, jump on too: https://t.co/bXghs4soMU

[cvberens @iamBrns]

@ayrondev_ @acgfbr @useblacksmith Using https://t.co/CqNRBYYgSx myself faster then Blacksmith and more reliable

[cvberens @iamBrns]

@shydev69 @Jitesh_117 Try avrea as well faster then Blacksmith

[cvberens @iamBrns]

@Zain_Wania @mscccc @depotdev @rbranson The job is to make the same workflow finish faster. If it’s running on a single core, parallelising the suite may be the first win. We can benchmark it on one deploy if useful. Try Avrea, same workflow, faster execution.

[cvberens @iamBrns]

@sa_mous @github Try https://t.co/CqNRBYYgSx no wait at all

[Avrea @avreacom]

Congratulations to Zero on their $10.3M seed round.

The Zero team ships fast and they ship with Avrea.

"Avrea promised a drop-in replacement and a 2-3x speedup, and that's what we got."
Olli Jokinen, Zero

https://t.co/rYhVhCRcRV

[cvberens @iamBrns]

@chrisreedbates @mikehostetler This is the kind of workflow we’re looking at with Avrea—giving agents room to work independently without getting stuck waiting on unrelated changes. Would love to learn more about your setup and see if we can help.

[cvberens @iamBrns]

@zekramu We are at Avrea try it out for free

[cvberens @iamBrns]

@igolaizola Avrea even better - EU based company with faster runners

[cvberens @iamBrns]

@gabriel_laet Make sense and we are building fast for that to be the case with our platform. More on that soon

[cvberens @iamBrns]

@0xblacklight You should try avrea if you wanna speed it up.

[cvberens @iamBrns]

@rpaweb This is really cool. Great idea to bring CI, deploy state and queue health into one place instead of stitching it together manually.
We’re thinking about something pretty similar at Avrea on the observability side. Would be interesting to compare notes sometime.

[cvberens @iamBrns]

@jetpen @0xblacklight This is the real bottleneck. Code generation is fast, but the remote build/test loop is still human-era slow.
Feels like build/test should move much closer to the agent environment, with CI becoming a thin verification layer.