[build-system]
requires = ["hatchling"]
build-backend = "hatchling.build"

[project]
name = "paritok"
version = "1.3.14"
description = "Open-source context compression for AI coding agents. A middle layer between your agent and the LLM API — self-host the 4B model or use the Paritok GPU server."
readme = "README.md"
license = "Apache-2.0"
requires-python = ">=3.10"
authors = [
    { name = "Jiayu Shi" },
    { name = "Luzhuo Chen" },
]
keywords = ["llm", "context", "compression", "coding-agent", "proxy", "token", "optimization", "ollama"]
classifiers = [
    "Development Status :: 4 - Beta",
    "Intended Audience :: Developers",
    "License :: OSI Approved :: Apache Software License",
    "Programming Language :: Python :: 3",
    "Topic :: Software Development :: Libraries",
    "Topic :: Scientific/Engineering :: Artificial Intelligence",
]
dependencies = [
    "tiktoken>=0.7.0",
    "click>=8.0",
    "pyyaml>=6.0",
    "httpx>=0.27.0",
]

[project.optional-dependencies]
dev = ["pytest", "ruff"]
# proxy ships the embedding tool-selector deps too, since tool_discovery defaults to
# "embedding". The bge-small model runs on CPU via fastembed/onnxruntime (no torch, no
# CUDA) and its quantized ONNX weights are BUNDLED in the wheel (paritok/data/bge-small-onnx/)
# — nothing downloads at runtime.
proxy = ["uvicorn>=0.30.0", "starlette>=0.38.0", "fastembed>=0.7", "numpy>=1.24"]
# Standalone embedding tool selection (engine without the proxy). CPU-only, free, no API,
# no torch — fastembed/onnxruntime over the bundled bge-small ONNX weights.
toolselect = ["fastembed>=0.7", "numpy>=1.24"]
# Persistent shadow storage (shadow_storage: redis) — [REF:id] entries survive restarts.
redis = ["redis>=5.0"]

[project.urls]
Homepage = "https://paritok.com"
Repository = "https://github.com/paritok-official/paritok-4b-v1"
Model = "https://huggingface.co/paritok/paritok-4b-v1"

[project.scripts]
paritok = "paritok.cli:main"

[tool.hatch.build.targets.wheel]
# Ship the whole package, including the verbatim training system-prompt data
# files under paritok/strategies/system_prompts/*.txt (loaded at runtime).
packages = ["paritok"]

[tool.ruff]
line-length = 100

[tool.ruff.lint]
# Pin the active rule set so a newer ruff release that turns on extra rules by
# default (0.16 began surfacing RUF/SIM/BLE) can't fail CI on unrelated code.
# E4/E7/E9 + F = pycodestyle errors + pyflakes — ruff's long-standing default.
select = ["E4", "E7", "E9", "F"]

[tool.ruff.lint.per-file-ignores]
# data_pipeline/ is one-off research/data tooling (its .jsonl output is
# gitignored and regenerable), so a few stylistic rules are relaxed there
# rather than churning throwaway code. The shipped paritok/ package stays strict.
"data_pipeline/**" = [
    "E741",  # ambiguous variable name (l/I/O) — fine in short analysis loops
    "E731",  # lambda assignment
    "E702",  # multiple statements on one line (semicolon)
    "F841",  # unused local (kept for readability in inspection scripts)
]