pyproject.toml

[project]
name = "inferhost"
version = "0.14.2"
description = "Self-hosted, multi-modal AI server for your own GPU: chat/vision LLMs (Qwen, Llama, Gemma, DeepSeek), text-to-speech (Kokoro, OuteTTS), and image generation (SDXL, Flux, Z-Image, Qwen-Image) behind one OpenAI-compatible endpoint. Wraps llama.cpp and stable-diffusion.cpp, auto-downloads GGUF models from Hugging Face, hot-swaps VRAM, speculative decoding — no compiling. Type `inferhost` and you're done."
readme = "README.md"
license = { file = "LICENSE" }
requires-python = ">=3.11"
authors = [
  { name = "amirrouh", email = "43451887+amirrouh@users.noreply.github.com" },
]
keywords = [
  "llm", "local-llm", "inference-server", "self-hosted", "local-ai",
  "openai-compatible", "openai-api", "gguf", "llama-cpp", "llama.cpp",
  "huggingface", "quantization", "speculative-decoding", "multimodal",
  "vision", "text-to-speech", "tts", "kokoro", "image-generation", "stable-diffusion",
  "sdxl", "flux", "qwen", "llama", "gemma", "gpu", "vulkan", "metal",
  "ai-server", "model-server", "tui",
]
classifiers = [
  "Development Status :: 3 - Alpha",
  "Environment :: Console",
  "Intended Audience :: Developers",
  "License :: OSI Approved :: Apache Software License",
  "Operating System :: POSIX :: Linux",
  "Operating System :: MacOS :: MacOS X",
  "Programming Language :: Python :: 3",
  "Programming Language :: Python :: 3.11",
  "Programming Language :: Python :: 3.12",
  "Programming Language :: Python :: 3.13",
  "Topic :: Scientific/Engineering :: Artificial Intelligence",
]

dependencies = [
  "textual>=0.80",
  "rich>=13",
  "huggingface-hub>=0.24",
  "httpx>=0.27",
  "litellm[proxy]>=1.40",
  # litellm's proxy imports fastapi.dependencies.utils.get_flat_dependant, which
  # FastAPI dropped in 0.140.7 (and every release after). litellm only asks for
  # fastapi<1.0, so a fresh install resolves past that and the gateway dies on
  # import — llama-swap keeps serving while port 4000 is refused. Last good
  # version is 0.140.6; unpin once litellm stops using the symbol.
  "fastapi<0.140.7",
  "pydantic-settings>=2",
  "psutil>=5.9",
  "pyyaml>=6",
  "tomli-w>=1",
  # Kokoro TTS engine (ONNX Runtime, CPU). Upstream caps python at <3.14, so
  # the marker keeps inferhost installable on newer pythons — Kokoro serving
  # raises a clear error there instead.
  "kokoro-onnx>=0.4.9; python_version < '3.14'",
]

[project.optional-dependencies]
# Deprecated alias: litellm is now a core dependency. Kept so 'inferhost[gateway]' still resolves.
gateway = []
dev = ["pytest>=8", "pytest-asyncio>=0.23", "ruff>=0.5", "mypy>=1.10"]

[project.scripts]
inferhost = "inferhost.cli:app"

[project.urls]
Homepage = "https://github.com/amirrouh/inferhost"
Issues = "https://github.com/amirrouh/inferhost/issues"

[build-system]
# Pin hatchling: newer releases emit Metadata-Version 2.5, which twine (in the
# PyPI publish action) rejects as "not a valid metadata version" and fails the
# release. 1.27.0 emits the widely-accepted 2.4. Bump only when the publish
# toolchain accepts 2.5.
requires = ["hatchling==1.27.0"]
build-backend = "hatchling.build"

[tool.hatch.build.targets.wheel]
packages = ["src/inferhost"]

[tool.ruff]
line-length = 100
target-version = "py311"

[tool.ruff.lint]
select = ["E", "F", "I", "UP", "B", "SIM"]
ignore = ["E501"]

[tool.pytest.ini_options]
testpaths = ["tests"]