Binary compression for typed APIs at the edge of entropy.
Typed APIs already know what their data can contain. Production traffic reveals what the data usually contains. Hyperfly uses both to compile a binary protocol for one exact route, instead of shipping generic JSON through a generic compressor.
Pre-release. The wire format is specified, three implementations agree on it byte-for-byte, and the benchmarks below are reproducible — but nothing is published and nothing is stable.
Bytes per message, averaged over 500-message corpora (bun run bench):
| route | JSON | JSON+Brotli | Protobuf | Hyperfly | + Brotli | Profiled |
|---|---|---|---|---|---|---|
| audit events | 12,687 | 2,512 | 7,190 | 2,013 | 1,974 | 574 |
| device telemetry | 7,994 | 1,422 | 2,007 | 755 | 725 | 542 |
| social feed | 6,863 | 2,294 | 4,396 | 1,871 | 1,868 | 1,466 |
| single order | 782 | 408 | 388 | 271 | 273 | 180 |
| OHLCV candles | 3,225 | 842 | 2,034 | 384 | 362 | 384 |
Read the spread rather than the best row. Training is worth 71% on audit logs — recurring values become dictionary codes, machine-made ids travel as grammar lanes, one column derives from another — and nothing at all on candles, whose only string sits outside the array. Note the Brotli column: on the profiled stream it now loses bytes on four of five routes. The corpora are synthetic — shaped like real routes, not captured from one — and no production traffic has been measured yet.
spec/ the authority: wire format, plans, negotiation, golden vectors
packages/hyperfly TypeScript reference implementation, zod adapter, HTTP layer
python/ Python implementation and pydantic adapter
rust/ Rust core
apps/interop a TS server and a Python client over real HTTP, run by CI
apps/bench corpora and the benchmark harness
apps/web hyperfly.dev
packages/lb legacy load balancer, previously published as `hyperfly@0.1.x`
The specifications are normative and the implementations are not. A fourth implementation ports against the golden vectors, not against this code.
- wire v0 — envelope, varints, bitmaps, node encodings, canonical artifacts, decoder limits
- plan columnar v5 — column layout, frame-of-reference bit packing, scaled-decimal and XOR numerics, packed text, trained dictionaries, grammar lanes, derived columns
- negotiation v1 — how peers agree on binary, how a client bootstraps, how a profile rotates without a cutover
bun install
bun run test # TypeScript
pytest python/tests -q # Python
cargo test --manifest-path rust/Cargo.toml
cd apps/bench && bun run benchCI runs all three suites, a Python version matrix, and the cross-language interop exchange on every push.