The RAG library that
works with everything

from cognity_ai import RAGLibrary

rag = RAGLibrary(
    llm="openai",
    embedder="openai",
    vector_store="qdrant",
    graph_store="neo4j",
    openai_api_key="sk-...",
    neo4j_password="your-password",
)

rag.ingest("contract.docx")
answer = rag.query("What are the payment terms?")

# 100% offline — no API keys, no cloud services
# pip install "cognity-ai[sentence-transformers,faiss,networkx,nlp]"
from cognity_ai import RAGLibrary

rag = RAGLibrary(
    llm="ollama",             # local Ollama server
    embedder="sentence_transformers",
    vector_store="faiss",
    graph_store="networkx",
    ocr="tesseract",
)

rag.ingest("local_docs/report.pdf")
answer = rag.query("Summarize the main points")

# Ingest ANY file type — format auto-detected
rag.ingest("report.pdf")          # PDF: page-aware, embedded images
rag.ingest("contract.docx")       # Word: headings, tables, images
rag.ingest("financials.xlsx")     # Excel: per-sheet text
rag.ingest("deck.pptx")           # PowerPoint: slides + notes
rag.ingest("data.csv")            # CSV: header-aware
rag.ingest("page.html")           # HTML: tag stripping
rag.ingest("photo.jpg")           # Image: Gemini Vision OCR
rag.ingest("scan.png")            # Image: complex layout OCR
rag.ingest("notes.md")            # Markdown: heading detection
rag.ingest("config.yaml")         # YAML/JSON: key-value text

# Or ingest an entire directory at once
results = rag.ingest_dir("./knowledge_base/", recursive=True)
print(len(results), "files ingested")

from cognity_ai.embedders.clip import CLIPEmbedder
from cognity_ai.ocr.whisper_local import WhisperLocalTranscriber
from cognity_ai.pipeline.video import VideoIngestionPipeline
from cognity_ai.retrievers.video import VideoRetriever

# Extract keyframes (visual) + transcribe speech
embedder     = CLIPEmbedder(model="openai/clip-vit-large-patch14")
transcriber  = WhisperLocalTranscriber(model_size="medium")

pipeline = VideoIngestionPipeline(
    embedder    =embedder,
    transcriber =transcriber,
    frame_interval=2,   # extract 1 keyframe every 2 seconds
)
pipeline.ingest("product_demo.mp4")

# Retrieve with timestamp-level precision
retriever = VideoRetriever(embedder=embedder, transcriber=transcriber)
results   = retriever.retrieve("product unboxing scene", top_k=3)
for r in results:
    print(f"[{r.timestamp_start:.1f}s – {r.timestamp_end:.1f}s] {r.transcript_snippet}")

from cognity_ai.ocr.whisper_local import WhisperLocalTranscriber
from cognity_ai.pipeline.audio import AudioIngestionPipeline
from cognity_ai.retrievers.audio import AudioRetriever

# Transcribe audio locally — no API keys required
transcriber = WhisperLocalTranscriber(model_size="large-v3", language="en")

pipeline = AudioIngestionPipeline(transcriber=transcriber)
pipeline.ingest("earnings_call.mp3")
pipeline.ingest("interview.wav")
pipeline.ingest_dir("./podcasts/")     # batch

# Retrieve transcript segments by meaning
retriever = AudioRetriever(transcriber=transcriber)
results   = retriever.retrieve("revenue guidance for next quarter", top_k=5)
for r in results:
    print(f"[{r.source}] score={r.score:.3f}")
    print(r.text)