systems that cannot crash.
catch cross-language bugs that single compilers miss.
curl -fsSL https://stokes.dev/install.sh | shrequires IBM Bob 2.0 · macOS · Linux · Windows (WSL)
works with
terminal
one command, three compilers.
Stokes scans your SQL schemas, Python pipelines, and Rust services in a single pass. It finds where they disagree and tells you exactly what broke.
benchmarks
fits entirely in L1 cache.
The hot path does partial selection in-place on a fixed stack buffer. No heap allocations, no latency spikes, all 200 feature descriptors in 25 cache lines.
Lower is better. Nanoseconds per packet feature partition.
How the hot path avoids panics when schemas grow.
case study
the $5.4M/hr single-compiler blind spot.
Based on the Cloudflare November 18, 2025 outage. Three compilers (sqlfluff, mypy, rustc) each passed on their own. But a schema change added 80 extra columns that overflowed a fixed-size buffer downstream, taking the whole fleet offline.
Internal replica tables quietly project 280 features instead of the expected 200 into KV metadata.
Automated KV sync pushes dynamic config fleet-wide. Early restarts look like transient jitter.
The proxy's fixed [Feature; 200] buffer gets 280 entries. Worker threads panic, edge traffic drops to zero.
Stokes partitions the buffer: 128 core rules stay pinned, 72 adaptive features get Zone 1, 80 extras get safely dropped. No heap allocations.
faq
frequently asked questions
common questions about how stokes works, what it checks, and how to run it in ci.
what is stokes?+
how does stokes prevent cloudflare-style outages?+
why does it run inside ibm bob 2.0?+
what do the benchmark numbers mean?+
what is the boundary lockfile?+
ready to verify?
install the cli, point it at your repo, and add `stokes verify` to your ci pipeline. takes about two minutes.