Introducing rill ⚡ A faster, drop-in replacement for Streamdown, the Markdown renderer for chat UIs
Built with Rust + WASM. Same React API, identical DOM output. Just swap the import
- 4.4× faster long-chat startup
- 6.6× less main-thread work
- 3× less memory
bun add @2018kguo/rill
Built with Rust + WASM. Same React API, identical DOM output. Just swap the import
- 4.4× faster long-chat startup
- 6.6× less main-thread work
- 3× less memory
bun add @2018kguo/rill
9
2
21
628
cranking a bunch of coding agents on your codebase = a massive CI bill. so I cut CI costs ~10x and made it much faster.
1/ switched from Blacksmith to boat.dev. ~13x cheaper compute per runner-minute, plus warm runners that cut setup from minutes to seconds.
2/ CPU-profiled every CI command. turns out TS loaders and spawning processes per test file cost more than the tests themselves. also cut redundant tests.
3/ stopped running all tests on every PR. static checks + a decision model (Jev) select relevant tests using the import graph. full suite still runs on a schedule.
4/ switched to faster tooling: tsrs/Bun check, oxlint, oxfmt, oxc loaders.
5/ forced Git to HTTP/1.1 to avoid HTTP/2 stream resets that occasionally made checkouts take 10+ minutes.
6/ don't forget the basics: cheap checks first, pre-commit hooks, path-based job filtering, cancel superseded runs, cross-run caching, and time-balanced test sharding.
wrote everything up as a skill:
npx skills add 2018kguo/optimize-ci
1/ switched from Blacksmith to boat.dev. ~13x cheaper compute per runner-minute, plus warm runners that cut setup from minutes to seconds.
2/ CPU-profiled every CI command. turns out TS loaders and spawning processes per test file cost more than the tests themselves. also cut redundant tests.
3/ stopped running all tests on every PR. static checks + a decision model (Jev) select relevant tests using the import graph. full suite still runs on a schedule.
4/ switched to faster tooling: tsrs/Bun check, oxlint, oxfmt, oxc loaders.
5/ forced Git to HTTP/1.1 to avoid HTTP/2 stream resets that occasionally made checkouts take 10+ minutes.
6/ don't forget the basics: cheap checks first, pre-commit hooks, path-based job filtering, cancel superseded runs, cross-run caching, and time-balanced test sharding.
wrote everything up as a skill:
npx skills add 2018kguo/optimize-ci
5
0
13
682
Right now it's around 20% more than Railway, but it depends on your workload.
There's a bunch you can do to bring it down like fargate spot for background jobs (~70% off), arm64/graviton (~20%), savings plans, NAT instances instead of NAT gateways if you push a lot of egress, or ECS on EC2 / plain EC2 for maybe another 10-30% if you're ok managing boxes.
We think we'll end up around Railway or a bit under
There's a bunch you can do to bring it down like fargate spot for background jobs (~70% off), arm64/graviton (~20%), savings plans, NAT instances instead of NAT gateways if you push a lot of egress, or ECS on EC2 / plain EC2 for maybe another 10-30% if you're ok managing boxes.
We think we'll end up around Railway or a bit under
1
0
0
227
Life update: after spending 3 years at @tryramp and building at @USTreasury, I started a new company with @shivamp! 🚀
At Ramp, the best memories came from hosting Ramp Hacks, rolling out our first AI PR reviewer on GPT 3.5, and building AI prototyping tools for the s. At Treasury, I built foundational infrastructure for the largest payments system in the world.
We’re building in one of the highest impact verticals with a team of future founders, researchers, and builders from top companies. More to come
We’ve already signed enterprise agreements and are growing our team in NYC. Reach out if interested!
(pic: day 1 of the company, out of blue bottle on university ave)
At Ramp, the best memories came from hosting Ramp Hacks, rolling out our first AI PR reviewer on GPT 3.5, and building AI prototyping tools for the s. At Treasury, I built foundational infrastructure for the largest payments system in the world.
We’re building in one of the highest impact verticals with a team of future founders, researchers, and builders from top companies. More to come
We’ve already signed enterprise agreements and are growing our team in NYC. Reach out if interested!
(pic: day 1 of the company, out of blue bottle on university ave)
87
17
289
32,639
Personal update, I started a company earlier this year with @rolandtshen! It’s been an amazing ride at @tryramp, joining as a founding member of the Workflows team when we were still in a WeWork, to seeing the company grow to $44B, 75,000 customers, and a team of 2000.
Grateful ships and for all the lessons I’ll take forward from seeing a generational company built on the inside: the bias towards constantly shipping, building an incredibly talent-dense team, and always scaling your ambition.
We’re tackling one of the largest verticals out there and are already helping early partners rebuild operational workflows in an AI-native way.
Our team today is ex-founders and staff engineers. We’re growing and looking for folks that thrive in an environment that optimizes for velocity. If this sounds exciting to you or a friend, please reach out!
Grateful ships and for all the lessons I’ll take forward from seeing a generational company built on the inside: the bias towards constantly shipping, building an incredibly talent-dense team, and always scaling your ambition.
We’re tackling one of the largest verticals out there and are already helping early partners rebuild operational workflows in an AI-native way.
Our team today is ex-founders and staff engineers. We’re growing and looking for folks that thrive in an environment that optimizes for velocity. If this sounds exciting to you or a friend, please reach out!
94
24
223
33,490
Today we’re launching Ramp Router.
3 years ago, we built an internal LLM router at @tryramp that powers AI products for 70,000 customers.
Back then it was mostly about saving money.
Now it feels obvious: the best model changes constantly. GPT, Claude, Gemini, Grok, Qwen, rices and capabilities move every week.
So we’re opening up access to everyone.
One OpenAI-compatible endpoint. The right model for every request. Lower cost without rewriting your app.
Reserve access to use it.
3 years ago, we built an internal LLM router at @tryramp that powers AI products for 70,000 customers.
Back then it was mostly about saving money.
Now it feels obvious: the best model changes constantly. GPT, Claude, Gemini, Grok, Qwen, rices and capabilities move every week.
So we’re opening up access to everyone.
One OpenAI-compatible endpoint. The right model for every request. Lower cost without rewriting your app.
Reserve access to use it.
215
201
1,816
893,056