Your GitHub Actions cache,
on your own hardware.

Your GitHub Actions cache,
on your own hardware.

A drop-in replacement for GitHub's cache service, written in Rust. actions/cache, setup-* caches and BuildKit type=gha work unchanged on self-hosted and ephemeral runners.

Get startedCopy agent promptGitHub★ 460

Ephemeral runners

Cache server

Your storage

• runner-7f3krunning job
• runner-2c9xrunning job
• runner-k41mrunning job
• unknown clientno valid token

github-actions-cache-server

1. verify runner tokenOIDC
2. scope to repo and branchisolated
3. look up key34k/s, p99 7 ms
4. save2.3 GiB/s
5. restore7.9 GiB/s

29 MB image0.5 s start3 MiB idle

• local NVMe
• S3, RustFS, MinIO, R2
• Google Cloud Storage
• Azure Blob

Restoring a 1 GiB cache62× faster than GitHub's cache from a self-hosted runner, on a 10 GbE network

1. GitHub cache, self-hosted runner~20 MB/s reported54 s
2. GitHub cache, GitHub-hosted runner~231 MB/s reported4.6 s
3. This server, over 10 GbEline rate of the link0.86 s
4. This server, same host7.9 GiB/s measured0.13 s

GitHub figures as reported by actuated (2023) and in google/wireit#239 (2022), your link and region decide the real number. Our method.

Two steps. Your workflows stay as they are.

1. 1Run the server

   docker-compose.yml

   services:
     cache-server:
       image: ghcr.io/falcondev-oss/github-actions-cache-server
       ports: ['3000:3000']
       environment:
         API_BASE_URL: http://cache-server:3000
       volumes: ['cache-data:/app/.data']
   volumes:
     cache-data:

2. 2Point your runners at it

   runner

   # your runner, ARC scale set or container
   image: ghcr.io/falcondev-oss/actions-runner
   env:
     CUSTOM_ACTIONS_RESULTS_URL: http://cache-server:3000/

   The forked runner keeps your cache URL and skips self-update. Prefer the stock image? Use the one-line patch.

3. 3There is no step 3

   .github/workflows/ci.yml (unchanged)

   - uses: actions/setup-node@v6
     with:
       cache: npm
   - uses: actions/cache@v4
     with:
       path: ~/.cargo
       key: cargo-${{ hashFiles('Cargo.lock') }}

Why not a mounted volume?

A hostPath only helps tools you point at it, on one node, shared by every repository. The cache server speaks GitHub's protocol instead.

Full comparison →

	GitHub cache	hostPath	This server
Cache stored	GitHub cloud	one node	your server
Workflow changes	none	every workflow	none
actions/cache, setup-*	yes	no	yes
Ephemeral runners	yes	same node only	yes
Restore speed	~20 MB/s	disk	LAN line rate
Per-repo isolation	yes	no	yes

Works with what you already run.

Anything built on @actions/cache with the v2 cache service.

Actions

actions/cacheactions/setup-nodeactions/setup-pythonactions/setup-goactions/setup-javadocker buildx type=ghaSwatinem/rust-cacheany @actions/cache tool

Storage

FilesystemS3, RustFS, MinIO, R2Google Cloud StorageAzure Blob

Database

SQLitePostgreSQLMySQL

Runs on

DockerHelmARCamd64, arm64GHES

Operations

Prometheus /metricsLRU evictionself-healing cleanupmanagement API

Let your agent set it up.

Every page is plain Markdown for agents. Append .md to any URL, or send Accept: text/markdown.

llms.txtllms-full.txtgetting-started.md

›Read https://gha-cache-server.falcondev.io/llms.txt and set up the GitHub Actions Cache Server for my self-hosted runners.Copy

Ship it in two minutes.

Get startedARC guideEphemeral runners with a local cache →

MIT licensed. Built by falconDev.