serviette
New Active Docs only Last checked 2026-11-16 Permissive (MIT/Apache/BSD) CPU OK Server / Docker
๐ฌ One weekly email: local RAG releases, hardware compatibility updates, and practical setup guides. Free. Unsubscribe anytime.
Updates on X: @mkjerman
What it is
serviette is a no-code RAG server from Pathway. You point it at a folder of documents, pick a vector database and an embedder in one YAML file, and get a chat UI and a REST API that answer with cited sources. Its selling point is that it stays live: per the README, edits, new files and deletions show up in answers within seconds, with no re-index step.
Per the README, local embeddings and the default embedded DuckDB store keep documents on your machine, and it also supports pgvector, Qdrant, Milvus, ChromaDB, Weaviate, Pinecone and MongoDB Atlas. Generated answers default to OpenAI; without a key it runs keyless and quotes the retrieved snippets. It needs a free Pathway license key. RAGShelf has not run it.
- License
- MIT (runs on the Pathway framework, which needs a free license key)
- Latest release
- v0.1.8
- Release date
- 2026-11-02
- Hardware
- Python server (pip install serviette); CPU OK with local embeddings; embedded DuckDB vector store by default, or 7 other vector DBs unverified
- Document types
- Text, Markdown, PDF, Office (DOCX, PPTX, XLSX, HTML, EML), scanned images via PaddleOCR; audio and video only with cloud API keys unverified
- OCR / tables / retrieval
- pypdf by default; optional Docling extra for layout-aware PDFs and tables; PaddleOCR extra for scans. Hybrid BM25 + vector, rerankers unverified
- Cloud calls
- Answer generation defaults to OpenAI (llm.type: openai); omit the key to run keyless (answers quote retrieved snippets). Embeddings can be local (serviette[local]). Needs a free Pathway license key (sign-in) unverified
- Fully offline
- Needs config
- GitHub stars
- 58,706
- Last commit
- 2026-11-10 (last push, GitHub API 2026-11-16)
Install
Copied from the official README. Check the project docs for options.
pip install "serviette[local]" --extra-index-url https://download.pytorch.org/whl/cpuserviette wizardserviette up --config config.yamlRetrieval
Vector search Hybrid (vector + keyword) Agentic / iterative
Based on: README: vector search over 8 vector DBs; optional hybrid BM25 + vector; adaptive RAG and query decomposition (checked 2026-11-16) (checked 2026-11-16). Vector vs. vectorless RAG explained โ
NVIDIA DGX Spark
DGX Spark: Unknown
No DGX Spark report found yet. We only mark a tool as working when there is a published source.
Last updated: 2026-11-16. License and release data from the GitHub API. Fields marked unverified have not yet been checked against official docs. How we know: Docs only (see the legend). Spot something wrong? Send a correction.
๐ฌ Get next week's tool
One new local RAG tool in your inbox every Monday. Free, unsubscribe anytime.
Sample issue: "AirLLM: run a 70B model on a 4GB GPU by loading one layer at a time. What it's good for (slow, private batch Q&A over your docs), what it isn't (fast chat), and the install command." Plus one new setup recipe or page, like our local vs. cloud cost calculator.