pyproject.toml

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

[project]
name = "auralynq"
version = "0.2.0"
description = "Auralynq — Talk to Your Data: a local-first, agentic, voice-enabled RAG platform with PathRAG graph retrieval."
readme = "README.md"
requires-python = ">=3.11"
license = { text = "Apache-2.0" }
authors = [{ name = "Auralynq" }]
keywords = ["rag", "pathrag", "voice", "agent", "langgraph", "qdrant", "retrieval"]
classifiers = [
  "Programming Language :: Python :: 3.11",
  "Programming Language :: Python :: 3.12",
  "License :: OSI Approved :: Apache Software License",
  "Operating System :: OS Independent",
]

# Core dependencies are intentionally lightweight and pure-Python where possible
# so the platform installs fast and the full test-suite runs offline at $0.
# Heavy / GPU / paid integrations live in optional extras and are imported lazily.
dependencies = [
  "pydantic>=2.7,<3",
  "pydantic-settings>=2.3,<3",
  "typer>=0.12,<1",
  "rich>=13.7",
  "httpx>=0.27",
  "numpy>=1.26",
  "networkx>=3.2",
  "structlog>=24.1",
  "tenacity>=8.2",
  "orjson>=3.10",
  "pyyaml>=6.0",
  "fastapi>=0.111",
  "uvicorn[standard]>=0.30",
  "sse-starlette>=2.1",
  "python-multipart>=0.0.9",
  "websockets>=12.0",
]

[project.optional-dependencies]
embeddings = ["sentence-transformers>=3.0", "FlagEmbedding>=1.2", "torch>=2.2"]
vector = ["qdrant-client>=1.9", "chromadb>=0.5"]
ingest = ["pypdf>=4.2", "python-docx>=1.1", "beautifulsoup4>=4.12", "markdown-it-py>=3.0", "pdfplumber>=0.11", "pymupdf>=1.24", "pdf2image>=1.17", "lxml>=5.0", "trafilatura>=1.8"]
# Cloud connectors. Notion + Slack use raw REST (httpx, already a core dep);
# only Google Drive's service-account auth needs the google SDK.
connectors = ["google-api-python-client>=2.0", "google-auth>=2.0"]
voice = [
  "faster-whisper>=1.0",
  "soundfile>=0.12",
  "librosa>=0.10",
  "silero-vad>=5.1",
]
agent = ["langgraph>=0.2", "langchain-core>=0.2"]
# Commercial LLM provider SDKs (optional; auto-detected, with offline fallback).
llm = ["openai>=1.30", "anthropic>=0.30", "cohere>=5.5"]
eval = ["ragas>=0.1.9", "jiwer>=3.0", "datasets>=2.19", "scikit-learn>=1.4"]
telemetry = [
  "opentelemetry-api>=1.25",
  "opentelemetry-sdk>=1.25",
  "opentelemetry-exporter-otlp>=1.25",
  "arize-phoenix>=4.0",
  "openinference-instrumentation>=0.1",
  "langfuse>=2.0",
]
# CPU/GPU local GGUF inference — fallback when Ollama is not running.
# Install the CUDA build for GPU: pip install llama-cpp-python --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cu121
slm = ["llama-cpp-python>=0.3", "huggingface_hub>=0.23"]
mcp = ["mcp>=1.0"]
# ColPali late-interaction visual retrieval (GPU-favored). torch comes from the
# `embeddings` extra; a deterministic offline hash fallback runs without this.
colpali = ["colpali-engine>=0.3", "pillow>=10.0"]
dev = [
  "ruff>=0.15.18,<0.16",
  "mypy>=1.10,<2.2",  # stability cap; type-check targets py3.12 (see [tool.mypy])
  "pytest>=8.2",
  "pytest-cov>=5.0",
  "pytest-asyncio>=0.23",
  "types-requests",
  "pre-commit>=3.7",
]
all = [
  "auralynq[embeddings,vector,ingest,voice,agent,llm,eval,telemetry,mcp]",
]

[project.scripts]
auralynq = "auralynq.cli:app"
auralynq-mcp = "auralynq.mcp_server.server:main"
auralynq-modelfit = "auralynq.modelfit.cli:main"

[project.urls]
Homepage = "https://github.com/MHHamdan/Auralynq"
Documentation = "https://github.com/MHHamdan/Auralynq#readme"

[tool.hatch.build.targets.wheel]
packages = ["auralynq"]

[tool.ruff]
line-length = 100
target-version = "py311"
src = ["auralynq", "tests", "scripts"]

[tool.ruff.lint]
select = ["E", "F", "I", "UP", "B", "C4", "SIM", "RUF", "TID"]
# UP042: str+Enum is an intentional, widely-used pattern (pydantic-friendly).
# RUF001: ambiguous unicode (curly quotes / en-dash) is intentional in entity
#         regexes and human-facing locators.
ignore = ["B008", "C901", "RUF012", "UP042", "RUF001"]

[tool.ruff.lint.per-file-ignores]
"tests/*" = ["B011", "SIM117"]
# planned.py embeds multi-line LLM prompt templates whose prose lines exceed
# the 100-col limit; reflowing them would alter the prompt text, so E501 is
# waived for this one file only.
"auralynq/rag/strategies/planned.py" = ["E501"]

[tool.mypy]
# Target 3.12: numpy>=2.4 ships PEP 695 `type` statements in its stubs, which
# mypy only accepts when both the target version and the interpreter running
# mypy are 3.12+. CI runs the type-check step on the 3.12 matrix job only.
python_version = "3.12"
warn_unused_ignores = false
warn_redundant_casts = true
disallow_untyped_defs = false
ignore_missing_imports = true
no_implicit_optional = true
check_untyped_defs = true
plugins = ["pydantic.mypy"]
exclude = ["web/", "build/", "tests/"]

[tool.pytest.ini_options]
asyncio_mode = "auto"
testpaths = ["tests"]
addopts = "-q --strict-markers"
markers = [
  "integration: requires optional heavy dependencies or network",
  "slow: long-running tests",
]
filterwarnings = ["ignore::DeprecationWarning"]

[tool.coverage.run]
source = ["auralynq"]
# Omit integration-only paths that require a live server / network / optional
# heavy dependency and therefore cannot run in the offline $0 unit suite. Their
# logic is covered by integration tests behind the `integration` marker.
omit = [
  "*/mcp_server/*",
  "*/__main__.py",
  "*/vectorstore/qdrant_store.py",
  "*/vectorstore/chroma_store.py",
  "*/retrieval/visual/colpali_embedder.py",
  "*/llm/providers.py",
  "*/llm/slm.py",
  "*/serving/worker.py",
  "*/embeddings/bge.py",
  "*/embeddings/ollama_embed.py",
  "*/modelfit/catalog_fetcher.py",
]

[tool.coverage.report]
show_missing = true
skip_covered = false