# Avrea

> Avrea is a drop-in replacement for GitHub Actions runners. 4th-gen AMD EPYC bare-metal hardware, local NVMe cache, and AI-assisted code quality. Builds run about 3x faster on cold runs and up to 145x faster with cache hits, at 33% lower per-minute pricing than GitHub-hosted runners. Free tier: 3,000 build minutes per month, no credit card.

Avrea is building the Software Delivery Platform for human-AI teams. The first product is live today: faster runners for GitHub Actions, adopted by changing one line in your workflow file. The roadmap covers workflow orchestration, AI-assisted quality, caching, and observability across the full delivery lifecycle.

Founded by Juha Valvanne (founder of Nosto) and Hannu Valtonen (founder of Aiven). Headquartered in Helsinki. Enterprise-ready: ISO 27001 certified, SOC 2 Type II attested, GDPR compliant.

### Key facts

- **Compatibility.** Drop-in for GitHub Actions runners. Keep your YAML, your secrets, your existing actions, and your `actions/cache`. GitLab CI, Bitbucket, and Jenkins are not supported.
- **Migration.** Change `runs-on: ubuntu-latest` to `runs-on: avrea-ubuntu-latest-2-vcpu` (or a different vCPU size). That is the entire migration.
- **Hardware.** AMD EPYC 4585PX at 5.7 GHz on Linux x86. Dedicated bare metal, not shared. Local NVMe at ~7 GB/s sequential read. 10 Gbps+ network. Fleet capacity 5,800+ vCPUs.
- **OS support.** Linux x86, Linux ARM, macOS, and Windows runners all live and generally available.
- **Speed.** About 3x faster on cold runs vs `ubuntu-latest`. Up to 145x faster with cache hits.
- **Real benchmark numbers.** Linux kernel build: 27m 23s on GitHub Actions, 24.1s on Avrea with cache (68.2x). Bazel: 37m 10s vs 56.1s (39.7x). Next.js + Turborepo: up to 145x.
- **Cache.** 25 GB quota per repo. Local NVMe, no network round trips. Works with `actions/cache`, Docker layer cache, and language package caches (npm, pip, Go, cargo).
- **Observability.** Per-step CPU and memory metrics, searchable logs, SSH into live jobs.
- **Pricing.** Free tier: 3,000 build minutes per month on 2 vCPU Linux x86 (1,500 minutes on 4 vCPU, 750 on 8 vCPU). Pay-as-you-go from $0.002/min (1 vCPU Linux x86) to $0.064/min (32 vCPU). Per-minute billing, no minimum. Enterprise tier available.
- **Compliance.** ISO 27001 certified. SOC 2 Type II attested. GDPR compliant.
- **Status.** Live and accepting customers. Free tier requires no credit card.

### Runner labels

All runners listed below are live. Linux x86, Linux ARM, macOS, and Windows are all generally available.

**Linux x86_64 (Ubuntu):**

| Label | vCPU | RAM |
|-------|------|-----|
| `avrea-ubuntu-latest-1-vcpu` | 1 | 4 GB |
| `avrea-ubuntu-latest-2-vcpu` | 2 | 8 GB |
| `avrea-ubuntu-latest-4-vcpu` | 4 | 16 GB |
| `avrea-ubuntu-latest-8-vcpu` | 8 | 32 GB |
| `avrea-ubuntu-latest-16-vcpu` | 16 | 64 GB |
| `avrea-ubuntu-latest-32-vcpu` | 32 | 128 GB |
| `avrea-ubuntu-latest-64-vcpu` | 64 | 256 GB |

Pinned-version variants: `avrea-ubuntu-24.04-N-vcpu` and `avrea-ubuntu-22.04-N-vcpu`.

**Linux ARM (Ubuntu):** `avrea-ubuntu-latest-arm-1-vcpu` through `avrea-ubuntu-latest-arm-16-vcpu` (1–16 vCPU, 4–64 GB RAM).

**macOS:** `avrea-macos-26-8-vcpu` and `avrea-macos-26-16-vcpu` (8–16 vCPU, 16–32 GB RAM).

**Windows:** `avrea-windows-2025-2-vcpu` through `avrea-windows-2025-16-vcpu` (2–16 vCPU, 8–64 GB RAM).

### How to migrate

```yaml
# Before
jobs:
  test:
    runs-on: ubuntu-latest

# After
jobs:
  test:
    runs-on: avrea-ubuntu-latest-2-vcpu
```

Same YAML, same secrets, same actions, same `actions/cache` API. No self-hosting, no agent install. Sign up at avrea.com, change the label, push.

### Six AI quality features

Tagline: "Your codebase gets better with every run." Full detail at https://avrea.com/quality.

1. **Failing test auto-fix.** Analyzes logs and test history to classify failures and write corrections.
2. **Pipeline optimization.** Identifies slow jobs, cache misses, and configuration inefficiencies, with proposed fixes.
3. **Dependency upgrades.** Tests version updates against the actual build and prioritizes security patches.
4. **Modern tooling.** Detects outdated build systems and handles safe migrations end-to-end.
5. **Risk-weighted coverage.** Combines test coverage with code change frequency to surface high-risk paths.
6. **Documentation sync.** Updates READMEs and guides when code changes make them outdated.

### What Avrea is not

- **Not a self-hosted runner.** Avrea runs on Avrea's bare-metal hardware. Nothing to install on your infrastructure.
- **Not a GitHub Actions replacement.** GitHub Actions still orchestrates the workflow. Avrea replaces only the runner.
- **Not available for GitLab CI, Bitbucket Pipelines, or Jenkins.** Those are roadmap items.
- **Not a shared-runner cloud.** Every job gets a dedicated VM. No noisy-neighbor effects, no shared-state flakes.
- **Not a generic AI coding assistant.** The AI features are scoped to CI: test fixes, pipeline tuning, dependency upgrades, coverage analysis.

### Public customers

Syncle, Luo, Focal, PromoRepublic, AceWaves, Saltfish, Keel.

Selected quotes (from avrea.com):
- Focal co-founder: "Avrea was 2x faster on top of that, and the remote cache is built into the runner image."
- Syncle CTO: "We reduced our job times by around 50% right away. We're especially excited about the AI quality features."
- PromoRepublic co-founder: "We've unlocked 4x faster execution and radically faster iteration."

## Documentation

- [docs.avrea.com](https://docs.avrea.com): Full product documentation, runner reference, configuration, and migration guides.

## Product pages

- [Speed](https://avrea.com/speed): Hardware specs, cache architecture, cold vs warm benchmark numbers, observable infrastructure detail.
- [Observability](https://avrea.com/observability): Per-step CPU and memory metrics, searchable logs, SSH debugging into live jobs.
- [Quality](https://avrea.com/quality): The six AI features with detail and examples. Tagline "Your codebase gets better with every run."
- [Pricing](https://avrea.com/pricing): Free tier definition, per-vCPU rates for all OS targets, Enterprise plan.

## Customer case studies

- [All case studies](https://avrea.com/case-studies): Index of customer write-ups.
- [Focal](https://avrea.com/case-studies/focal): 2x on top of cache, remote cache built into the runner image.
- [Luo](https://avrea.com/case-studies/luo): Customer story.
- [PromoRepublic](https://avrea.com/case-studies/promorepublic): 4x faster execution, faster iteration cycles.
- [Syncle](https://avrea.com/case-studies/syncle): Job times cut ~50%, AI quality features in production.

## Benchmarks and engineering posts

- [Linux kernel CI benchmark](https://avrea.com/blog/linux-kernel-ci-benchmark): Linux kernel build, 68.2x faster with cache (24.1s vs 27m 23s on GitHub-hosted).
- [Bazel CI benchmark](https://avrea.com/blog/bazel-ci-benchmark): 39.7x faster with cache (56.1s vs 37m 10s).
- [Next.js + Turborepo CI benchmark](https://avrea.com/blog/nextjs-turborepo-ci-benchmark): Up to 145x faster with cache.
- [Kafka + Gradle CI benchmark](https://avrea.com/blog/kafka-gradle-ci-benchmark): Apache Kafka build comparison.
- [Ghostty + Nix CI benchmark](https://avrea.com/blog/ghostty-nix-ci-benchmark): Nix flake build for the Ghostty terminal emulator.
- [RisingWave + sccache CI benchmark](https://avrea.com/blog/risingwave-sccache-ci-benchmark): Rust build with sccache.
- [How we made GitHub Actions cache up to 6x faster](https://avrea.com/blog/how-we-made-github-actions-cache-up-to-6x-faster): Cache architecture deep-dive.

## Trust and compliance

- [Trust and security](https://avrea.com/trust-security): ISO 27001, SOC 2 Type II, GDPR posture, security practices, responsible disclosure.
- [Avrea is SOC 2 Type 2 attested](https://avrea.com/blog/avrea-is-soc-2-type-2-attested): SOC 2 Type II attestation announcement.
- [Avrea is ISO 27001 certified](https://avrea.com/blog/avrea-iso-27001-certified): ISO 27001 certification announcement.
- [Service level agreement](https://avrea.com/sla): SLA terms.
- [Service description](https://avrea.com/service-description): Formal product description.

## Company

- [About Avrea](https://avrea.com/about-us): Mission, founders, approach.
- [Careers](https://avrea.com/careers): Open roles, Helsinki-based team.
- [Open application](https://avrea.com/open-positions/open-application): General application for unlisted roles.
- [Hannu Valtonen](https://avrea.com/team/hannu-valtonen): Co-founder. Previously co-founded Aiven.
- [Jeppe Tiensuu](https://avrea.com/team/jeppe-tiensuu): Team page.
- [Leo Lännenmäki](https://avrea.com/team/leo-lannenmaki): Team page.

## Get in touch

- [Book a demo](https://avrea.com/demo): Demo request form.
- [Schedule a meeting](https://avrea.com/meeting-scheduler): Direct calendar booking.
- [Contact](https://avrea.com/contact): General contact.

## Blog

- [Blog](https://avrea.com/blog): All posts, including benchmarks, engineering deep-dives, and company announcements.

## Optional

- [Privacy policy](https://avrea.com/privacy): Data handling and privacy practices.
- [Terms of service](https://avrea.com/terms): Master service terms.
- [Payment terms](https://avrea.com/payment-terms): Billing and payment terms.
- [Information for consumers](https://avrea.com/information-for-consumers): Required consumer information.