OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
RAG & Knowledge Systems

Verbatim Rag

Verbatim Rag is a Python library for retrieval-augmented question answering that retrieves documents, selects relevant passages, and returns source excerpts with citations. It includes a full ingestion, indexing, retrieval, extraction, and response pipeline, plus a separate lightweight core for transforming a question and supplied context into evidence.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Verbatim Rag — OpenViST project profile in RAG & Knowledge Systems.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use it when an application needs to expose the source text behind an answer, or when you want an extractive step between retrieval and response presentation. Choose the full library for ingestion and retrieval, or verbatim-core when context is already supplied.

Who it is for

  • Developers building knowledge-base question-answering systems that need cited source passages.
  • Research teams creating document workflows where readers should inspect the evidence selected for an answer.
  • Application builders who want a provenance-focused retrieval layer rather than unrestricted answer rewriting.

Documented uses

  • Process and chunk documents for a searchable collection.
  • Index documents and retrieve relevant material in response to a question.
  • Extract answer-relevant spans and present them as source excerpts with citations.
  • Transform already-retrieved context into an evidence response using the lightweight core.
Before you choose

Requirements & limits

The repository identifies the project as MIT-licensed Python software. The full package is installed as verbatim-rag; the lightweight core is installed separately as verbatim-core.

  • The full pipeline uses Docling and Chonkie for document conversion and chunking, with vector indexing and retrieval.
  • Extraction can use a fine-tuned BERT-based model or an OpenAI-based LLM path; the README also describes local encoder configurations that avoid generative LLM API calls.
  • Static template mode is identified for fixed response framing; the default contextual template may generate surrounding presentation text.
  • The PyPI wheel does not include the repository's FastAPI API and Vite/React development UI.
  • Provenance covers returned evidence text on built-in exact/static paths, not source truth, retrieval recall, passage relevance, or completeness.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Build a local retrieval and cited-answer pipeline

Process a PDF URL, index it with sparse SPLADE retrieval, and query for an answer passage.

Before you start

  • Python environment with verbatim-rag and its pipeline dependencies installed
  • A PDF URL and document title
  • Local Milvus store
  • SPLADE model weights; the README notes that model downloads require network access

Steps

  1. Use DocumentProcessor.process_url to process the PDF URL with a title and optional metadata.

  2. Create SpladeProvider with the documented sparse model name and device setting.

  3. Create LocalMilvusStore with sparse indexing enabled, then pass it and the provider to VerbatimIndex and add the processed document.

  4. Wrap the index in VerbatimRAG, call query with a question, and inspect the returned answer.

Expected outcome

A query response containing selected source text and citations to indexed documents.

Guide limitations

  • Retrieval can miss relevant documents, and extraction can return irrelevant or incomplete passages.
  • The README's quick-start setup also documents an OpenAI API-key environment variable; local configurations without generative API calls are described separately.
  • Provenance does not establish source truth or answer completeness.
Documentation 1 ↗
Source-based practical guide

Transform supplied context with the lightweight core

Turn a question and already available text passages into an evidence response without setting up the full retrieval pipeline.

Before you start

  • Python environment with verbatim-core installed
  • A question and context entries containing content and, optionally, titles

Steps

  1. Install the documented verbatim-core package.

  2. Import VerbatimTransform and create an instance.

  3. Call transform with the question and a list of context objects.

  4. Read the response answer.

Expected outcome

An evidence response derived from the supplied context, using the core transform rather than document ingestion and retrieval.

Guide limitations

  • This lightweight core does not include the full RAG pipeline.
  • The README lists openai, pydantic, rapidfuzz, and jinja2 as dependencies.
Documentation 1 ↗
COMPARE OPTIONS

Compare options for Retrieval-augmented applications

These projects document a common task and share a project role. This is a source-based comparison, not an execution benchmark or a guarantee of interchangeability.

ProjectDocumented purposePlatformAccess and requirements
Verbatim RagProcess and chunk documents for a searchable collection.CLI, Linux, macOS, WindowsThe repository identifies the project as MIT-licensed Python software. The full package is installed as verbatim-rag; the lightweight core is installed separately as verbatim-core.
LangsearchCrawl website or local data for a knowledge base.CLI, Self-hosted, Server
Hybrid Search RagCombine document retrieval and language-model generation in a configurable workflow.
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
13 October 2026
Original website title
KRLabsOrg/verbatim-rag · Official homepage title · retained in its source language
How we review sourcesSuggest a correction