Audited 2026-10-20 · Sources: Cloudflare engineering blog post, GitHub repo metadata + file tree, npm listing. Tabletop audit — nothing installed or executed.
The card
| Repo | oxc-project/tsrs |
| Stars | 1.1k (GitHub, audit date) — 1,797 commits, 7 issues / 9 PRs. PRs are collaborator-only right now. |
| License | Apache-2.0 |
| Provenance | Cloudflare + VoidZero + the Oxc project + Owner.com's Max Schwenk, working in close collaboration with the TypeScript team itself. Explicitly not a fork — the TS team keeps full control of the language. |
| What it is | A Rust port of the TypeScript 7 type checker, incubating inside the Oxc toolchain. Ships as npm i tsrs / npx tsrs plus a WASM build (@tsrs/wasm) that runs in browsers and on Workers. |
| Install surface | npm package — standard toolchain slot. It's a type checker: reads your code, emits diagnostics. No daemon, no service. |
What problem it solves
Type checking is the slowest part of a modern JS build. tsrs claims up to 8x faster and ~70% less memory than TypeScript 7.0 at the same worker count — and the WASM target means a full type-checking toolchain can run inside a browser tab or an edge sandbox, not just your laptop.
Our gate findings
| Gate | Finding |
|---|---|
| Upstream identity | Verified — official oxc-project org, announced on the Cloudflare blog, co-developed with VoidZero, with named engineers and a public collaboration statement from the TS team's side. As provenance goes, this is the ceiling. |
| License | Apache-2.0 — clean, patent grant included. |
| Claims discipline | Excellent. The 8x/70% numbers were benchmarked with the TypeScript team's help — same configuration, not the bigger-machine tricks other ports have used. They also disclose that some optimizations were already upstreamed into TS 7.1 beta (which alone is ~2.5x over 7.0). Honest benchmarking is rare enough to name. |
| Maturity | Experimental, their word. Diagnostics print order still differs from upstream; merging with pingdotgg's tsc-rs for LSP and content mappers is in-flight. Do not gate CI on it yet — treat as a preview. |
| Maintenance model | Worth paying attention to: they keep parity with upstream TypeScript using the Vinext agent pattern — every morning an agent reviews upstream diffs and files tracking issues; every night a compatibility matrix regenerates. The repo even carries .agents/skills and tasks/ directories. The toolchain is maintained by the same kind of fleet we run. |
| Blast radius | Low to try (npx tsrs touches nothing but the code you point it at). The strategic weight is bigger: type-checking inside WASM means dev environments — and agent sandboxes — can run the full TS pipeline where it physically couldn't before. |
| Data boundary | Clean — a compiler is a local batch tool; WASM build runs client-side or in your own Worker. |
Verdict: GO — for experimentation. Provenance, license, and benchmark honesty all clear. The only caution is maturity: it's labeled experimental for a reason. Run it on side projects, watch the diagnostics drift, and let it earn the build pipeline slot later.
Best for
Teams whose CI burns real minutes on type checking, and anyone building browser/edge-side dev tooling — the WASM story is the part that changes what's buildable, not just what's faster.