Java, Rust, and .NET clients reach feature parity
Six client libraries (TypeScript, Python, Java, Rust, .NET, C) are now byte-identical across the cross-stack benchmark matrix. 36 cells × 3 sizes, all green.
The Codec polyglot client matrix is feature-complete. Every client (TypeScript via @codecai/web, Python via codecai, Java, Rust, .NET, and C) ships:
- Frame decoder (msgpack + protobuf)
- Detokenizer (byte_level + metaspace + byte_fallback)
- BPETokenizer (deterministic, bit-identical to HuggingFace’s reference)
- ToolWatcher (region detection without decoding)
- Translator (cross-vocab agent handoff)
- MapLoader (sha256-verified, well-known discovery)
The cross-stack benchmark matrix runs all six clients against all three text engines (sglang, vLLM, llama.cpp) at three prompt sizes — 6 × 3 × 3 = 54 cells, all byte-identical. A single tokenizer-map registry; one wire shape; six languages.