RAG

RAG combines document loading, embedding, retrieval, and answer generation.

Create a RAG instance

rag = akasha.RAG(
    model="gemini:gemini-2.5-flash",
    embeddings="gemini:gemini-embedding-001",
    chunk_size=1000,
    search_type="auto",
)

Common constructor options:

Option	Meaning
model	Chat model used to generate the answer.
embeddings	Embedding model used for documents and queries.
chunk_size	Approximate size of document chunks.
search_type	Retrieval strategy, commonly auto.
max_input_tokens	Maximum input size for the final answer.
use_chroma	Use an existing Chroma-backed data source when applicable.
stream	Whether answer generation is streamed.

Ask about documents

import akasha

rag = akasha.RAG(
    model="gemini:gemini-2.5-flash",
    embeddings="gemini:gemini-embedding-001",
)

answer = rag(
    data_source="./docs",
    prompt="What are the main ideas in these documents?",
)
print(answer)

Call shape:

answer = rag(
    data_source=["notes.md", "report.pdf"],
    prompt="Summarize the important findings.",
)

The normal return value is a final str. The first run may take longer because documents and embeddings need to be prepared.