voice-rag 0.1.0

Provider-agnostic voice RAG pipeline. Plug in your voice provider, LLM, vector store, and document parsers.

voice-rag

voice-rag is a Python package for voice-oriented RAG pipelines: ingest local documents, store them in Qdrant, serve an OpenAI-style chat-completions webhook, and swap voice or LLM providers behind a stable interface.

What it includes

• Python SDK via KnowledgeAgent
• CLI for init, ingest, serve, query, inspect, and doctor
• Built-in voice adapters for elevenlabs and deepgram
• Built-in LLM clients for openai, anthropic, and gemini
• Built-in parsers for .txt, .md, .pdf, and .docx
• Qdrant vector store integration with hybrid dense + BM25 retrieval

Quickstart

pip install "voice-rag[elevenlabs]"
export OPENAI_API_KEY=your_api_key_here
voice-rag init

Edit voice-rag.yaml, then ingest and serve:

voice-rag ingest ./data/sample_docs/claude-code-changelog.md --recreate
voice-rag serve

Point your voice platform to http://localhost:8000/v1 if it expects an OpenAI-style Custom LLM endpoint. By default, voice-rag uses an embedded local Qdrant store in .qdrant, so you do not need to start a separate Qdrant server unless you set vector_store.url.

CLI

voice-rag init [--dir PATH]
voice-rag ingest <path> [--recreate] [--config PATH]
voice-rag serve [--host HOST] [--port PORT] [--reload] [--config PATH]
voice-rag query <text> [--limit N] [--config PATH]
voice-rag inspect [--config PATH]
voice-rag doctor

Python API

from voice_rag import KnowledgeAgent, VoiceRagConfig

config = VoiceRagConfig()
agent = KnowledgeAgent(config=config)
agent.ingest("./docs", recreate=True)
app = agent.create_app()

Configuration

voice-rag reads configuration from voice-rag.yaml and environment variables. It does not auto-load .env files.

Key	Env	Default
llm.api_key / embedding.api_key	OPENAI_API_KEY	(required for OpenAI)
llm.provider	LLM_PROVIDER	openai
llm.model	LLM_MODEL	gpt-4o-mini
embedding.model	EMBEDDING_MODEL	text-embedding-3-small
vector_store.url	VECTOR_STORE_URL	empty; use local .qdrant
vector_store.collection_name	VECTOR_STORE_COLLECTION_NAME	knowledge_base
vector_store.local_path	VECTOR_STORE_LOCAL_PATH	.qdrant
server.port	SERVER_PORT	8000

See voice-rag.yaml for the full schema.

Development

pip install -e ".[all,dev]"
pytest tests/ -v

Use CONTRIBUTING.md for connector and packaging guidelines.

Publishing to PyPI

With PYPI_TOKEN and the release environment configured in the repo, push a version tag to trigger the GitHub Actions workflow:

git tag v0.1.0
git push origin v0.1.0

The workflow runs smoke test → build → publish to PyPI.