L

langgraph-rag

Free

Evidence-grounded RAG platform combining hybrid retrieval, reranking, citation verification, and fail-closed answer generation.

Visit Website
generative-aiagentic-aiai-reliabilitybm25citation-verificationaiai assistantautomation

Overview

EvidenceFlow is a LangGraph-based agentic RAG system for research and document intelligence, built around sparse-first retrieval, evidence provenance, citation verification, and fail-closed answers. It is presented as a trust-oriented pipeline rather than a conventional vector-search chatbot.

Its retrieval architecture is deliberately sparse-first and vector-database-free. The knowledge-base retrieval layer does not depend on dense embeddings or k-NN vector search; instead it uses OpenSearch's inverted index and BM25 for classical lexical retrieval, with optional OpenSearch neural-sparse retrieval. If neural-sparse model deployment is unavailable, the system falls back to the classical lexical path.

Retrieval starts with a bounded, deterministic first pass. When that pass is weak, an LLM planner generates a small number of alternative lexical queries sent back through the same OpenSearch retrieval layer, bounded by a hard round limit. Retrieved candidates are combined with Reciprocal Rank Fusion and can be passed to a cross-encoder reranker, which is an enhancement rather than a hard dependency.

The trust layer treats retrieved documents and web results as untrusted evidence, isolates retrieved content from agent control flow, and validates citations against the actual evidence set used for synthesis. It distinguishes states such as evidence verified, limited evidence, conflicting evidence, unable to verify, and unsafe content blocked.

Pricing

Free

Detailed plans have not been confirmed in our catalog. Check the official website for current limits and billing terms.

Visit Website

Prices and limits may change. Confirm the currency, billing period, seat minimum and usage caps on the official website.

Use Cases

  • Teams use EvidenceFlow to answer questions over a knowledge base with a vectorless lexical retrieval path.
  • Teams use EvidenceFlow to combine knowledge-base and web research in a single agentic pipeline.
  • Teams use EvidenceFlow to check that generated answers cite the evidence that actually supports them.
  • Teams use EvidenceFlow to surface conflicting evidence and claims involving numbers and dates.
  • Teams use EvidenceFlow to refuse unsupported answers when evidence is insufficient or conflicting.
  • Teams use EvidenceFlow to improve retrieval ranking with RRF fusion and cross-encoder reranking.
  • Teams use EvidenceFlow to run a containerized deployment with OpenSearch, optional web research, and configurable Jina reranking.

Features

Hybrid information retrieval

Retrieved result reranking

Citation verification for answers

Fail-closed answer generation

Evidence-grounded RAG responses

Reviews

No reviews yet. Be the first to share your experience!

Rankings & collections