ragxpy 0.1.0

Production-grade RAG in 4 lines — hybrid search, streaming, and agent tools on by default.

ragxpy

The pip-installable Python RAG library built for 2026 — hybrid search, streaming, and agent tools on by default.

No Docker. No background server. No framework lock-in. Just Python.

Docs · Changelog · Contributing · Discussions

Install

pip install ragxpy

Quickstart

from ragxpy import RAG

async with RAG(llm="openai/gpt-4o-mini") as rag:
    await rag.ingest("./docs/")
    answer = await rag.query("What is the refund policy?")
    print(answer.text)
    print(answer.citations)

Why ragxpy?

Feature	ragxpy	LangChain	LlamaIndex	RAGFlow
Lines to get started	4	40+	20+	Docker setup
Hybrid search on by default	✓	✗	opt-in	✓ (Docker)
pip install (no server)	✓	✓	✓	✗
Async-first	✓	partial	partial	✗
Streaming responses	✓	partial	partial	✗
Agent tool built-in	✓	✗	✗	✗
Multi-tenant isolation	✓	✗	✗	✗
Incremental indexing	✓	✗	opt-in	✗
Built-in eval	✓	✗	partial	✗

Chunking accuracy: RecursiveChunker matches FloTorch Feb 2026 benchmark winner at 69% end-to-end accuracy.

Who It's For

✓ Python developers who want production-ready RAG as a library, not a platform.
✓ Teams already on PostgreSQL — zero new infrastructure with store="postgresql://...".
✓ Anyone who values typed, async-first, minimal-dependency code.

✗ Not for you if you need a no-code UI, knowledge graphs, or agent orchestration — use RAGFlow or LangGraph instead.

Store Options

# Development — volatile, zero setup
RAG(store="memory")

# Development — persistent across restarts, embedded
RAG(store="lance://./ragxpy-index")

# Production — PostgreSQL + pgvector, native hybrid search
RAG(store="postgresql://user:pass@localhost/mydb")

Upgrade Path

store="memory"                  # local dev / tests
    ↓
store="lance://./ragxpy-index"  # persistent dev, survives restarts
    ↓
store="postgresql://..."        # production (pgvector benchmark: 471 QPS @ 50M vectors)

Configuration

from ragxpy import RAG, QueryConfig

async with RAG(
    embedder="openai/text-embedding-3-small",
    store="lance://./my-index",
    llm="openai/gpt-4o-mini",
    chunk_size=512,
    chunk_overlap=64,
    cache=True,
) as rag:
    await rag.ingest("./docs/", glob="**/*.md")
    answer = await rag.query(
        "What changed in v2?",
        config=QueryConfig(top_k=5, include_citations=True),
    )

Optional Extras

pip install ragxpy[lance]       # LanceDB persistent store
pip install ragxpy[postgres]    # PostgreSQL + pgvector
pip install ragxpy[local-emb]   # sentence-transformers embedder + reranker
pip install ragxpy[eval]        # RAGAS + Langfuse eval loop
pip install ragxpy[serve]       # ragxpy serve HTTP API

CLI

ragxpy init          # generate ragxpy_config.py with defaults
ragxpy serve         # start HTTP API on localhost:8000
ragxpy serve --port 9000

Streaming

async for token in rag.stream_query("What changed in v2?"):
    print(token, end="", flush=True)

Works with OpenAI, Anthropic, and Ollama. Falls back gracefully for custom LLMs.

Agent tools

from ragxpy.agent import as_claude_tool

# Give your Claude agent access to your document index
tool = as_claude_tool(rag)              # Anthropic-compatible tool schema
results = await rag.search("query", top_k=5)  # raw retrieval, no generation

Pass as_openai_tool(rag) for OpenAI function calling.

Multi-tenant isolation

# Tag documents by tenant at ingest time
await rag.ingest("./org_a_docs/", tenant_id="org_a")
await rag.ingest("./org_b_docs/", tenant_id="org_b")

# Filter by tenant at query time — no code changes to the store
answer = await rag.query("policy?", config=QueryConfig(tenant_id="org_a"))

Roadmap

v0.1.0 is the foundation. Follow GitHub Discussions for what's coming next — roadmap is shaped by real usage and feedback.

Community

• Questions & help → GitHub Discussions
• Bug reports → GitHub Issues
• Contributing → CONTRIBUTING.md

License

MIT — see LICENSE