hush-providers

LLM, embedding, and reranking provider integrations for Hush workflows.

Installation

pip install hush-providers

Quick Start

LLM (chain = prompt + LLM combined)

from hush.core import Hush, GraphOp, START, END, PARENT
from hush.providers import chain

async def main():
    with GraphOp(name="chat") as graph:
        chat = chain(
            resource="gpt-4o",
            template={"system": "You are a helpful assistant.", "user": "{question}"},
            question=PARENT["question"],
        )
        START >> chat >> END

    result = await Hush(graph).run(inputs={"question": "What is Python?"})
    print(result["content"])

Embedding

from hush.providers import EmbeddingOp

embed = EmbeddingOp.of(resource="bge-m3", texts=PARENT["documents"])

Reranking

from hush.providers import RerankOp

rerank = RerankOp.of(resource="bge-reranker", query=PARENT["query"], documents=PARENT["docs"])

Supported Providers

Configuration

Providers are configured via YAML resource files:

# resources.yaml
llm:
  gpt-4o:
    type: openai
    model: gpt-4o
    api_key: ${OPENAI_API_KEY}

embeddings:
  bge-m3:
    type: onnx
    model_path: /models/bge-m3

Feature Flags

Install only the providers you need:

pip install "hush-providers[openai]"       # OpenAI + Azure
pip install "hush-providers[gemini]"       # Google Gemini
pip install "hush-providers[onnx]"         # ONNX Runtime
pip install "hush-providers[all-light]"    # All without PyTorch
pip install "hush-providers[all]"          # Everything

Rust Backend

All providers have native Rust implementations via hush-providers (crate) — direct HTTP calls without Python overhead.

Related Packages

License

Apache 2.0