Same build.ninja format
rninja parses the exact build.ninja your generator emits — CMake, GN, Meson, or hand-written. It does not regenerate the file; it executes it, preserving phony, restat, pools, and depfiles.
Home / Features
Everything ninja does, plus the parts it leaves out
rninja keeps the format and the CLI and closes the gaps in the executor: a content-addressed cache, an async scheduler, and optional cross-machine sharing. Here is the full breakdown.
Drop-in compatibility
rninja is ninja where ninja works. Same format, same flags, nothing to migrate.
Mirrors ninja core flags
The flags you already type work: -C, -f, -j, -k, -l, -n, -t, -v, -d. NINJA_STATUS is honored and exit codes match, so scripts and CI do not move.
All of ninja -t, plus extras
clean, compdb, graph, deps, and query behave like ninja. rninja adds -t config for inspecting or generating an rninja.toml and a stats view for cache hit rate.
Content-addressed caching
Hash every action over its real inputs. A hit is a guaranteed-equivalent artifact.
Local action cache
Every action is hashed over its inputs, command line, and observed environment. Identical inputs produce identical hashes, so repeat builds skip the toolchain entirely.
Crash-safe sled index
The cache index is backed by sled, so an interrupted build cannot corrupt the store. Restart and pick up where you left off.
Strictly additive correctness
A miss falls through to the original ninja action; a hit only happens when inputs, command, and environment all match. Caching never changes what a build produces.
Scheduling
Keep every core busy where stock ninja stalls on single-threaded bookkeeping.
Tokio-driven async executor
The scheduler moves bookkeeping off the hot path and keeps cores saturated, so the misses you do have to build finish sooner.
IO-aware ready queue
The ready queue accounts for IO contention, not just raw job count, so a graph that is disk-bound in places does not thrash the CPU.
Remote cache
Optional. Share content-addressed artifacts across laptops and CI runners.
async-nng transport
The optional remote cache streams blobs between machines over async-nng. One small endpoint can serve a whole team.
Fresh machines start warm
First builds on a new laptop or a clean CI runner become mostly cached builds once a peer has seeded the store with identical inputs.
Opt-in, local-first
Enabled via RNINJA_REMOTE_URL and RNINJA_CACHE_MODE. Leave it off and rninja is a purely local cache plus scheduler — the local store always front-runs the remote.
Install & operation
No daemon to register, no service to run. Sensible defaults out of the box.
Four package channels
Ships on crates.io (cargo install rninja), npm (rninja-cli), PyPI (rninja-cli), and Homebrew. Prebuilt binaries are on GitHub Releases.
Zero-config defaults
The cache lives under $XDG_CACHE_HOME/rninja. No daemon to install. RNINJA_* env vars exist for power users; everyone else gets the speedup for free.
MIT licensed
A memory-safe Rust implementation, MIT licensed, built by Neul Labs.
See it on your own build
Read how it works, follow the quickstart, or check the comparisons.