Projects from the community. If you've built something with NexRAG, open a GitHub issue to get it listed here.
A Streamlit chatbot that answers questions about company policy, grounded in your Markdown/text policy docs. Streams answers token-by-token via stream_query(), ships OpenTelemetry + Prometheus observability and LLM-as-judge evaluation, and runs fully local on Ollama — every pipeline decision lives in YAML, with no custom RAG code.
Query a corpus of legal contracts for specific clauses, obligations, and dates. Ingest PDFs, ask in plain English, get structured answers with source citations.
An always-accurate assistant over internal documentation. Ingests Confluence exports, Notion pages, and Markdown files. Employees ask questions, get cited answers.
Load a folder of academic PDFs, embed with a local HuggingFace model, retrieve with a strict score threshold. Cross-paper synthesis with zero API costs.
Index product docs and past support tickets. Surface relevant context to agents in real time. Fully local with Ollama — no customer data leaves the org.
Point NexRAG at a repo's Markdown docs and source comments. Ask architecture questions, get cited answers. The built-in code chunker preserves function boundaries.
FastAPI backend serving SSE token streams via astream_query(). Ingestion runs async with parallel embedding batches. Users see answers appear word-by-word as the LLM generates them.
Open a GitHub issue with your project name, description, stack, and a link. We'll add it here.
Any problem where the answer lives in your documents — not in the model's training data.
PDFs, Word docs, plain text — ingest any document corpus and query it in natural language. Sources and scores are always returned.
Give teams an always-accurate assistant over internal docs. Ingest incrementally. Deduplication handles re-ingestion safely.
Load academic papers, reports, or structured data. Run local embeddings with Ollama — no API costs, no data leaving your machine.
Use NexRAG as the retrieval backbone for any LLM-powered product. Control exactly what context the model sees, and verify it with citations.
Set mode: async and wire astream_query() into a FastAPI SSE endpoint. Tokens arrive live from the LLM — no buffering, no polling.