better-auth/better-auth
How Better Auth got 2× faster E2E tests and cut 20,000 CI minutes a month with StarSling.
In the weeks after migrating to StarSling Runners, StarSling agents shipped three optimization PRs and cut per-job E2E runtime from 2m 22s to just over one minute, adding up to roughly 20,000 minutes of CI compute saved every month.
“We've been using StarSling for Better Auth CI. It's been saving us countless hours, both with fast runners and their agent optimizing our CI.”
- 2×
- Weekly CI runs
~1,800 → ~4,000 / week
Seed·$5M raised·San Francisco
Intro
Better Auth is the most popular open-source TypeScript authentication framework with over 28,000 stars on GitHub, over 800 contributors, and hundreds of PRs open at any given time. Better Auth is scaling fast: since turning on StarSling on February 11, 2026, weekly npm downloads have gone from 1.2 million to 3.3 million, and their CI load has roughly doubled alongside it going from 1,800 workflow runs per week to close to 4,000 workflow runs a week. A growing framework means a growing test suite, and CI work that perpetually competes with shipping the framework itself.
Problem
A five-minute E2E tax, about to be paid twice as often
At migration, end-to-end tests took close to five minutes on every PR, a tax the growing contributor queue was about to pay twice as often. The bottlenecks were the kind that only compound as a framework grows: a turbo cache that wasn't picking up hits between runs, a Docker Compose stack whose services weren't ready when the suite kicked off, and Playwright reinstalling browsers from scratch on every run.
“Before StarSling, speeding up CI was the sort of work that didn't make it into a release. There was always a feature that mattered more than shaving a minute off the test suite. Now we can actually ship CI improvements with new releases.”
Solution
CI that opens its own optimization PRs
Better Auth turned on self-driving CI on February 11. Over the next nine days StarSling agents opened three more PRs on their own: fix the turbo cache configuration, add Docker Compose healthchecks where services were racing the test runner, and cache Playwright browser installs across runs. Three CI-tuning projects shipped in nine days.
“StarSling agents are like a CI engineer we never had to hire. They find the slow spots, test out fixes, find the ones that work and open the PRs themselves.”
Results
2m 22s → 1m 04s
In the weeks after migration, once the runner swap and all three agent-authored optimizations had landed, typical per-job E2E runtime measured 2m 22s → 1m 04s (2.22×) and CI 1m 40s → 1m 02s. Across the thousands of CI and E2E runs Better Auth fires off every month, those per-job savings compound to roughly 20,000 minutes of CI compute saved. Not one of those optimizations was scoped by a Better Auth engineer: the agents found the work, opened the PRs, and the team reviewed and merged. For a library every team's PR queue gates on, that's the durable win: CI that improves itself instead of sitting in the backlog behind the framework's own roadmap.
“Better Auth is growing fast. We've almost doubled our GitHub Actions usage in the past few months without CI becoming a bottleneck thanks to StarSling.”
Timeline
When each speedup landed
Migration on 2026-02-11. Every dot is a merged starsling/* PR. Hover for the title.
2026-02-11
2026-02-20
- #79332026-02-11migration
chore: migrate ci workflows to StarSling runners
- #79502026-02-13
chore: fix turbo cache configuration in ci
- #80102026-02-16
chore(ci): add Docker Compose healthchecks for faster CI service readiness
- #80732026-02-20
chore(ci): optimize Playwright browser installs in E2E
← Migration
Workflow speedups
Where the time went
Measured at p50 job-duration across all branches. Pre-migration: GitHub-hosted runners. Post: StarSling runners. Dates in the methodology note below.
E2E
2.22×Methodology
1Per-job p50 runtime: the median duration of an individual CI job, across all branches. Before: GitHub-hosted runners, the 3 days before the migration PR merged. After: StarSling runners with the agents' three optimization PRs merged, April 1 to 4.
2CI compute minutes, the unit GitHub Actions bills: each job's measured p50 saving (CI ~38s, E2E ~78s) summed across every parallel job, times Better Auth's actual monthly job volume from the GitHub Actions API. A conservative, success-runs-only estimate.
“Most CI providers just give you faster machines. StarSling gives us that and made our tests faster, so we could stay focused on making Better Auth even better.”