Sono migrated from Depot to Avrea in 15 minutes achieving 50% faster CI performance

An already optimized voice-AI pipeline experienced measurable performance improvements on Avrea’s Apple M5 Max ARM runners, requiring only a minimal change to existing workflow configurations.

A blue background with a green and white pattern.

Results

Depot to Avrea migration: about 15 minutes, a couple of pull requests
Standard runner build time reduced 50% (prior to cache optimization)
Supabase startup step: 2x faster
ARM integration suite on Apple M5 Max: consistent performance gains
Workflow changes: Updated runner labels on GitHub Actions

‍

About Sono

Sono (callsono.com) develops AI-powered voice assistants for service businesses. These systems manage high-volume inbound and outbound calls, including booking, routing, confirmations, and follow-ups. Sono’s team consists of senior engineers who have had their share of dealing with CI pains before.

Technical Stack‍

- GitHub Actions
- Supabase/Postgres
- ARM-based infrastructure
- End-to-end testing suites simulating real AI-driven calls

Prior to adopting Avrea, Sono utilized Depot and GitHub Actions for CI, alongside AWS Batch for ad hoc workloads.

CI speed was already a priority

Sono’s CI pipeline was already highly optimized, with parallelized builds and multi-core execution. The system supported both continuous build-and-test workflows and end-to-end test suites involving live AI call simulations.

Given this baseline, any performance gains required improvements beyond an already efficient system.

Migration process

Avrea migration was easy: sign in with GitHub, point the runner labels in your workflow files at Avrea. Avrea runs ARM on Apple M5 Max, so Sono had the hardware they wanted in the same shape they already knew.

Sono moved the pipeline over in fifteen minutes, across a couple of PRs. Most caching came on without extra setup. Sono had a high bar going in: Sono's CI was already optimized, so any gain had to come on top of a tuned baseline.

Performance Improvements

On a standard runner, a build that took three minutes on Depot ran in two on Avrea, and that was before caching was tuned for the repo. The Supabase startup step came up noticeably faster. On the ARM integration suite, which is mostly sleep and timeout waiting by design, there were still clear speedups.

Faster ARM CI in one place gives them room to grow that test volume, plus a path to pull in work that had been split out to AWS Batch. And because the migration cost no project time, none of it traded against shipping.

"Waiting on CI drives me crazy, so our pipeline was already optimized. Avrea was still clearly faster, and the switch from Depot took a couple of simple PRs."

— Elias Nygrén, Co-founder & CTO, Sono

Elias's take for other teams: if you are on GitHub Actions, there is no reason not to try it.

On GitHub Actions and/or Depot today? Try it out for free at console.avrea.com