Cache
Shared Build Cache
Make completed build work available to the next machine. Reuse native tool results and directory caches across CI and local development without changing runner providers.
Completed build work
- Compiler results
- Dependencies and task outputs
- Docker layers
Hosted CI
Self-hosted
Local builds
Choose the tool you already use.
Connect a native cache or save a directory. Keep the build system that decides what can be reused.
Docker
Layers, cache mounts, and compiler results 1.89× faster than GHA CacheCargo / Rust
Dependencies, target state, and sccache 1.28× faster than GHA Cache + targetBazel
Action results and build outputs 1.49× faster than BuildBuddyGradle
Cacheable task outputsMaven
Build Cache Extension results 1.40× faster than Depot CacheNx
Task results for your monorepoTurborepo
Shared task outputsccache
C and C++ compiler results 1.11× faster than GHA Cachesccache
Rust and C/C++ compiler resultsGo
Build and cacheable test results 2.19× faster than Depot CacheNix
Completed store paths and closuresXcode
Native compilation cache for macOSBuild time by tool
Average restore, build and save time for one public project per tool, on the same runner class and commits as other caches.
PostHog
Average restore, build and save · lower is better
GitHub runner
4 vCPU · 16 GB · Ubuntu 24.04 x86-64
Base · Image layers
Plus · Image layers, tool caches and cache mounts
- Saved per build
- 13 min 20 s
- Faster than GHA Cache
-
1.89×
1.96× on the 10 build pairs with the same CPU model
- Builds compared
- 30
Depot runner
4 vCPU · 16 GB · Ubuntu 24.04 x86-64
Base · Image layers
Plus · Image layers, tool caches and cache mounts
Depot runner
8 vCPU · 32 GB · Ubuntu 24.04 x86-64
Base · Image layers
Plus · Image layers, tool caches and cache mounts
Depot remote builder
16 vCPU · 32 GB* · x86-64
* Builder sizes published by providers.
- Saved per build
- 36.0 s
- Faster than Depot builder, 16 vCPU
-
1.10×
BoringCache on Depot runner, 8 vCPU: 5 min 58 s against 6 min 34 s on the builds both ran
- Builds compared
- 30
Namespace runner
4 vCPU · 16 GB · x86-64
Base · Image layers
Plus · Image layers, tool caches and cache mounts
Namespace remote builder
8 vCPU · 16 GiB* · x86-64
* Builder sizes published by providers.
- Saved per build
- 47.5 s
- Faster than Namespace builder, 8 vCPU
-
1.08×
BoringCache on Namespace runner, 4 vCPU: 9 min 16 s against 10 min 3 s on the builds both ran
- Builds compared
- 18
Zed
Average restore, build and save · lower is better
GitHub runner
4 vCPU · 16 GB · Ubuntu 24.04 x86-64
Base · sccache compiler cache
Plus · sccache, Cargo registry and target directory
- Saved per build
- 8 min 33 s
- Faster than GHA Cache + target
-
1.28×
1.21× on the 6 build pairs with the same CPU model
- Builds compared
- 30
Depot runner · Zed release bundle
8 vCPU · 32 GB · Ubuntu 24.04 x86-64
- Saved per build
- 6 min 38 s
- Faster than Depot runner cache
-
1.51×
1.48× on the 29 build pairs with the same CPU model
- Builds compared
- 30
Namespace runner with 16 GB swap · Zed on Namespace runners
4 vCPU · 16 GB · Ubuntu 24.04 x86-64
Base · sccache compiler cache
Plus · sccache, Cargo registry and target directory
Namespace runner with cache volume · Zed on Namespace runners
4 vCPU · 16 GB · Ubuntu 24.04 x86-64
These runners have no other cache at the same level to compare with BoringCache.
gRPC
Average restore, build and save · lower is better
GitHub runner
4 vCPU · 16 GB · Ubuntu 24.04 x86-64
- Saved per build
- 83.2 s
- Faster than BuildBuddy
-
1.49×
1.33× on the 1 build pair with the same CPU model
- Builds compared
- 9
Hugo
Average restore, build and save · lower is better
GitHub runner
4 vCPU · 16 GB · Ubuntu 24.04 x86-64
- Saved per build
- 49.7 s
- Faster than Depot Cache
-
2.19×
2.26× on the 7 build pairs with the same CPU model
- Builds compared
- 18
Lichess
Average restore, build and save · lower is better
GitHub runner
4 vCPU · 16 GB · Ubuntu 24.04 x86-64
- Saved per build
- 2 min 59 s
- Faster than GHA Cache
-
2.52×
1.75× on the 3 build pairs with the same CPU model
- Builds compared
- 9
Spring AI
Average restore, build and save · lower is better
GitHub runner
4 vCPU · 16 GB · Ubuntu 24.04 x86-64
- Saved per build
- 95.3 s
- Faster than Depot Cache
-
1.40×
1.30× on the 7 build pairs with the same CPU model
- Builds compared
- 17
OBS Studio
Average restore, build and save · lower is better
GitHub runner
4 vCPU · 16 GB · Ubuntu 26.04 x86-64
- Saved per build
- 11.8 s
- Faster than GHA Cache
-
1.11×
1.28× on the 11 build pairs with the same CPU model
- Builds compared
- 28
All published projects, runners and caches are on the benchmark page. See all benchmarks →
Restore dependencies in seconds. Send less when they change.
BoringCache restored a 2.7 GB dependency cache in 10.0–17.1 seconds across four CI providers. Saving it after a dependency change sent about 90% less data than the compared caches.
| Runner | Compared cache | BoringCache restore | Other restore |
|---|---|---|---|
| GitHub | GitHub Actions cache | 17.1 s | 17.2 s |
| Depot | Depot runner cache | 11.5 s | 5.7 s |
| Namespace | GitHub Actions cache | 11.9 s | 18.7 s |
| Blacksmith | Blacksmith cache | 10.0 s | 15.5 s |
| Runner / cache | Initial save | Initial egress | Reuse save | Reuse egress |
|---|---|---|---|---|
| GitHub / BoringCache | 15.7 s | 117 MB | 9.5 s | 76 MB |
| GitHub / Actions cache | 14.1 s | 757 MB | 17.3 s | 757 MB |
| Depot / BoringCache | 8.2 s | 117 MB | 6.0 s | 76 MB |
| Depot / Depot runner cache | 10.9 s | 756 MB | 10.5 s | 759 MB |
| Namespace / BoringCache | 9.6 s | 114 MB | 6.8 s | 72 MB |
| Namespace / Actions cache | 7.3 s | 743 MB | 7.4 s | 727 MB |
| Blacksmith / BoringCache | 10.6 s | 113 MB | 9.5 s | 72 MB |
| Blacksmith / Blacksmith cache | 6.1 s | 722 MB | 7.0 s | 723 MB |
One run per cache and runner. Saves use an existing workspace: the initial save uses a new key, and the reuse save follows a dependency change. Egress is data sent during the save.
Upload the changes. Keep less duplicated data.
Matching cache content is stored once within a workspace. Related builds can reuse existing data instead of uploading and storing another full copy.
One workspace
Cache entry 1
Cache entry 2
Shared cache. Fast reads on the build machine.
Local serving and preloading help supported tools read cached results with less waiting. Data missing locally is fetched from the workspace.
Bring reusable state to fresh runners and keep local data between builds on persistent machines.
Inspect reuse and transfers →2.7 ms
Median cache read time
Median of tool-reported project averages across 3 benchmarks. Hits can be served locally after prefetch.
Your build tools
Native cache integrations
BoringCache CLI
Local hit → serve cached data
Fetch data missing locally
Shared workspace cache
Docker build cache
A changed Docker layer need not mean a cold build.
Share layers, then add package and compiler caches for work inside instructions that have to run again.
Explore Docker caching →Docker layers
Reuse an unchanged instruction's output.
Compiler and tool caches
Reuse matching results when an instruction runs again.
Persistent cache mounts
Keep package downloads and build directories between runs.
1.89×
Faster Docker builds
Best Docker result against the GitHub Actions cache on the same runner class.
Choose who can contribute to shared cache.
Reuse does not have to mean giving every job write access. Keep access scoped to the workspaces each job needs.
Explore security →Separate readers and publishers
- Trusted CI jobs Publish shared cache for the next build Read + publish
- Pull requests Reuse cache without replacing it Read only
- Sandboxes and agents Reuse cache and keep new outputs local Read only
Connect supported CI through OIDC instead of keeping a reusable BoringCache secret in your workflow.
Try the next build with shared cache.
Start with one tool in one repository. Try it for 30 days. No credit card required.
curl -sSL https://install.boringcache.com/install.sh | sh cd your-project boringcache onboard