Cofone

Simple, fast, yours. Turn documents, websites and videos into a queryable knowledge base in a few lines of Python.

from dotenv import load_dotenv
from cofone import RAG

load_dotenv()  # reads OPENROUTER_API_KEY from .env

answer = RAG().add_source("docs/").run("Who is Leonardo?")
print(answer)

What is Cofone?

Cofone is an open-source Python RAG (Retrieval-Augmented Generation) library.
Load any document, ask questions in natural language, get precise answers — without complex setup or boilerplate.

Key highlights:

• Fluent DSL — chain everything in one expression
• BM25 (default) + FAISS semantic search
• 19 LLM providers: OpenRouter, OpenAI, Anthropic, Gemini, Mistral, Groq, Cohere, DeepSeek, xAI, Together, Perplexity, Fireworks, Cerebras, NVIDIA, DeepInfra, Anyscale, Ollama, LM Studio, llama.cpp
• 10 embedding providers: local sentence-transformers, OpenAI, Gemini, Cohere, Mistral, Voyage, Jina, NVIDIA, Together, Ollama
• Smart chunking that respects document structure
• Chat memory, streaming, structured output (Pydantic)
• FAISS index persistence to disk
• Sources: files, folders, PDFs, web URLs, Wikipedia, YouTube

Installation

pip install cofone

With optional extras:

pip install "cofone[pdf]"      # PDF support (pypdf)
pip install "cofone[faiss]"    # FAISS semantic search (faiss-cpu + sentence-transformers)
pip install "cofone[web]"      # Wikipedia + YouTube
pip install "cofone[all]"      # everything above

Setup — API key required

Cofone needs at least one LLM provider API key.
The default provider is OpenRouter — free tier available, 200+ models, one key.

Step 1: Get a free key at openrouter.ai/keys

Step 2: Create a .env file in your project folder:

OPENROUTER_API_KEY=sk-or-...

Step 3: Load it in your script:

from dotenv import load_dotenv
load_dotenv()

Or pass the key directly (no .env needed):

RAG(model_api_key="sk-or-...").add_source("docs/").run("question")

→ Full setup guide for all providers: INSTALL.md

Examples

from dotenv import load_dotenv
from cofone import RAG
load_dotenv()

# ── Sources ───────────────────────────────────────────────────────────────────

# single file
RAG().add_source("notes.txt").run("Summarize")

# folder — loads all .txt .md .pdf recursively
RAG().add_source("docs/").run("What is the main topic?")

# PDF (requires pip install "cofone[pdf]")
RAG().add_source("report.pdf").run("What are the conclusions?")

# Wikipedia
RAG().add_source("https://en.wikipedia.org/wiki/Python").run("What is Python?")

# YouTube transcript
RAG().add_source("https://www.youtube.com/watch?v=VIDEO_ID").run("Summarize this video")

# multiple sources combined
RAG().add_source("docs/").add_source("https://en.wikipedia.org/wiki/AI").run("Overview")

# ── LLM providers ─────────────────────────────────────────────────────────────

RAG(model_provider="openai",     model="gpt-4o-mini").add_source("docs/").run("question")
RAG(model_provider="anthropic",  model="claude-3-5-haiku-20241022").add_source("docs/").run("question")
RAG(model_provider="gemini",     model="gemini-2.0-flash").add_source("docs/").run("question")
RAG(model_provider="groq",       model="llama-3.1-8b-instant").add_source("docs/").run("question")
RAG(model_provider="ollama",     model="llama3").add_source("docs/").run("question")  # local, no key

# ── FAISS semantic search ─────────────────────────────────────────────────────

# local embeddings (no extra key)
RAG(faiss=True).add_source("docs/").run("Find concepts related to learning")

# OpenAI embeddings
RAG(faiss=True,
    embedding_provider="openai",
    embedding_model="text-embedding-3-small"
).add_source("docs/").run("question")

# fully local — Ollama LLM + Ollama embeddings, no internet, no keys
RAG(model_provider="ollama",    model="llama3",
    faiss=True,
    embedding_provider="ollama", embedding_model="nomic-embed-text"
).add_source("docs/").run("question")

# ── Chat memory ───────────────────────────────────────────────────────────────

bot = RAG().add_source("docs/")
bot.chat("Who is Leonardo da Vinci?")
bot.chat("When was he born?")          # knows the context — "he" = Leonardo
bot.chat("What are his best works?")

# custom system prompt — tell the LLM how to behave
RAG(
    system_prompt="You are an art historian. Answer only about paintings and sculptures."
).add_source("docs/").run("Tell me about Leonardo")

# ── Streaming ─────────────────────────────────────────────────────────────────

for token in RAG().add_source("docs/").stream("Tell me about this document"):
    print(token, end="", flush=True)
print()

# ── Structured output (Pydantic) ──────────────────────────────────────────────

from pydantic import BaseModel

class Person(BaseModel):
    name: str
    birth_year: int
    nationality: str

data = RAG().add_source("docs/").run("Extract data about Leonardo", schema=Person)
print(data.name)        # Leonardo da Vinci
print(data.birth_year)  # 1452

LLM Providers (19 total)

Embedding Providers (10 total)

Links

• 📦 PyPI: pypi.org/project/cofone
• 💻 GitHub: github.com/LeonardoCofone/cofone
• 📖 Full feature reference: FEATURE.md
• 🔧 Installation guide: INSTALL.md

License

MIT — see LICENSE