v0.2.0

AGON v0.2.0

What's Changed

• Feat/rust bindings @Verdenroz (#6)
  • Full Changelog: v0.1.0...v0.2.0

Supported platforms:

• Linux: x86_64, aarch64 (manylinux)
• macOS: x86_64 (Intel), aarch64 (Apple Silicon)
• Windows: x64
• Python: 3.11, 3.12, 3.13

Rust Core Implementation

The entire encoding/decoding engine has been rewritten in Rust (~4,000 lines) and exposed via PyO3:

• AGONRows: Row-based tabular encoding
• AGONColumns: Columnar transpose encoding
• AGONStruct: Template-based encoding
• Parallel format evaluation with rayon
• Fast byte-length token estimation (opt-in tiktoken)

Format Rename: AGONText → AGONRows

from agon import AGON, AGONRows

# New (recommended)
result = AGON.encode(data, format="rows")

Token Counting Options

# Fast estimation (default, uses len/4 heuristic)
result = AGON.encode(data, format="auto")

# Accurate tiktoken counting (slower but precise)
result = AGON.encode(data, format="auto", encoding="o200k_base")

Documentation

• Full docs: https://verdenroz.github.io/agon-python/
• API Reference: https://verdenroz.github.io/agon-python/api/
• Benchmarks: https://verdenroz.github.io/agon-python/benchmarks/

For Contributors

Development now requires the Rust toolchain:

# Install Rust
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh

# Build and test
uv sync --dev
make build
uv run pytest

Full Changelog

Added

• Rust core implementation with PyO3 bindings (crates/agon-core/)
• Parallel format evaluation using rayon
• Fast byte-length token estimation as default
• Cross-platform wheel builds (Linux, macOS, Windows)
• Comprehensive performance benchmarks in documentation

Changed

• Renamed AGONText → AGONRows throughout codebase and docs
• Moved Python code from src/ to python/
• Updated CI/CD for Rust + Python hybrid builds
• Pre-commit hooks now include Rust linting