README.md

---
title: Auralynq RAG
emoji: 🎙️
colorFrom: blue
colorTo: purple
sdk: docker
app_port: 7860
pinned: false
license: apache-2.0
short_description: "Voice-native agentic RAG on Llama-3.3-70B"
---

# Auralynq RAG

[Auralynq](https://github.com/MHHamdan/Auralynq) is a local-first, voice-native,
agentic RAG platform with span-level visual citation grounding, a PathRAG
knowledge graph, a Compounding Wiki, and an observable retrieval pipeline.

**This Space** runs the single-container build (FastAPI API + Next.js UI on one
port) configured to use **`meta-llama/Llama-3.3-70B-Instruct`** through Hugging
Face **Inference Providers** (the OpenAI-compatible router), so answers are
generated by a powerful hosted model — no local GPU required.

## Configuration

Set as Space **Secrets / Variables** (applied by the deploy step):

| Key | Value | Kind |
|---|---|---|
| `HUGGINGFACE_TOKEN` | *your HF token* | secret |
| `AURALYNQ_LLM__PROVIDER` | `huggingface` | variable |
| `AURALYNQ_LLM__MODEL` | `meta-llama/Llama-3.3-70B-Instruct` | variable |
| `AURALYNQ_DEMO_MODE` | `true` (seeds the demo corpus on first boot) | variable |
| `AURALYNQ_ALLOW_UPLOADS` | `false` (uploads disabled) | variable |

Retrieval uses hash embeddings + an in-memory vector store in this build (no
model downloads); switch to real embeddings by upgrading the image extras and
setting `AURALYNQ_EMBEDDING__PROVIDER`.

> Generation calls are billed to the token owner's Hugging Face account. This
> Space is **private** by default — keep it private unless you intend to let
> others spend your Inference Providers credits.

Source & docs: <https://github.com/MHHamdan/Auralynq>