retrieval·agent
An autonomous retrieval-augmented generation agent.
Plug in any vector store, any LLM, any reranker. Hybrid search, reranking, query rewriting, an LLM quality gate, and an autonomous retry loop — built on LangGraph.
from retrievalagent import init_agent
rag = init_agent("documents", model="openai:gpt-5.4", backend="qdrant")
state = rag.chat("What is the status of operation overlord?")
print(state.answer) # Cited. Grounded. Delivered.
True Agentic Loop
Retrieve → judge → rewrite → retry. Fully autonomous up to max_iter rounds. Never satisfied with good enough.
Lean Parallel Pipeline
prepare → [keyword ∥ synonym] → quality_gate → merge_rerank → generate. Keyword and synonym searches fan out in parallel; semantic backup fires only when BM25 scores low.
Synonym + Spell-Correct
One LLM call per query: synonym expansion, spell-correction (as a parallel BM25 term, not a query overwrite), and negation extraction ("cola nicht zero" → excluded_terms=["zero"]).
Hybrid Search
BM25 + vector search fused with Reciprocal Rank Fusion or Distribution-Based Score Fusion. MMR diversity pass removes near-duplicate docs before reranking.
8 Backends
Meilisearch, Azure AI Search, ChromaDB, LanceDB, Qdrant, pgvector, DuckDB, InMemory. Swap with one line.
Multi-Reranker
Cohere, HuggingFace, Jina, ColBERT, RankGPT, embed-anything, or any custom reranker.
HyDE
Hypothetical Document Embeddings boost recall on vague or descriptive queries automatically.
Auto Strategy
LLM samples your collection at init, tunes hyde_style_hint, semantic_ratio, and domain hints — zero per-query overhead.
Tool-Calling Agent
invoke_agent gives the LLM a toolset: inspect schema, sample field values, build filters on the fly, boost by business signals.
Multi-Collection
Route queries to the right subset of collections automatically. LLM picks which indexes to search per request.
Install
pip install retrievalagent[recommended] # Meilisearch + Cohere + CLI
pip install retrievalagent # Base only — InMemory backend
pip install retrievalagent[all] # Everything
See Quick Start for all install options and a 5-minute walkthrough.