- Swap LLMs anytime – Switch between 100+ LLMs via LiteLLM or run local models.
- Type-safe LLM calls – Use Python generics to enforce strict type safety in model interactions.
- Bring your own vector store – Connect to Qdrant, PgVector, and more with built-in support.
- Developer tools included – Manage vector stores, query pipelines, and test prompts from your terminal.
- Modular installation – Install only what you need, reducing dependencies and improving performance.
- Ingest 20+ formats – Process PDFs, HTML, spreadsheets, presentations, and more. Process data using Docling, Unstructured or create a custom parser.
- Handle complex data – Extract tables, images, and structured content with built-in VLMs support.
- Connect to any data source – Use prebuilt connectors for S3, GCS, Azure, or implement your own.
- Scale ingestion – Process large datasets quickly with Ray-based parallel processing.
- Real-time observability – Track performance with OpenTelemetry and CLI insights.
- Built-in testing – Validate prompts with promptfoo before deployment.
- Auto-optimization – Continuously evaluate and refine model performance.
- Chat UI – Deploy chatbot interface with API, persistance and user feedback.
To get started quickly, you can install with:
pip install ragbitsThis is a starter bundle of packages, containing:
ragbits-core- fundamental tools for working with prompts, LLMs and vector databases.ragbits-agents- abstractions for building agentic systems.ragbits-document-search- retrieval and ingestion piplines for knowledge bases.ragbits-evaluate- unified evaluation framework for Ragbits components.ragbits-chat- full-stack infrastructure for building conversational AI applications.ragbits-cli-ragbitsshell command for interacting with Ragbits components.
Alternatively, you can use individual components of the stack by installing their respective packages.
To define a prompt and run LLM:
import asyncio
from pydantic import BaseModel
from ragbits.core.llms import LiteLLM
from ragbits.core.prompt import Prompt
class QuestionAnswerPromptInput(BaseModel):
question: str
class QuestionAnswerPromptOutput(BaseModel):
answer: str
class QuestionAnswerPrompt(Prompt[QuestionAnswerPromptInput, QuestionAnswerPromptOutput]):
system_prompt = """
You are a question answering agent. Answer the question to the best of your ability.
"""
user_prompt = """
Question: {{ question }}
"""
llm = LiteLLM(model_name="gpt-4.1-nano", use_structured_output=True)
async def main() -> None:
prompt = QuestionAnswerPrompt(QuestionAnswerPromptInput(question="What are high memory and low memory on linux?"))
response = await llm.generate(prompt)
print(response.answer)
if __name__ == "__main__":
asyncio.run(main())To build and query a simple vector store index:
import asyncio
from ragbits.core.embeddings import LiteLLMEmbedder
from ragbits.core.vector_stores import InMemoryVectorStore
from ragbits.document_search import DocumentSearch
embedder = LiteLLMEmbedder(model_name="text-embedding-3-small")
vector_store = InMemoryVectorStore(embedder=embedder)
document_search = DocumentSearch(vector_store=vector_store)
async def run() -> None:
await document_search.ingest("web://https://arxiv.org/pdf/1706.03762")
result = await document_search.search("What are the key findings presented in this paper?")
print(result)
if __name__ == "__main__":
asyncio.run(run())To build a simple RAG pipeline:
import asyncio
from pydantic import BaseModel
from ragbits.core.embeddings import LiteLLMEmbedder
from ragbits.core.llms import LiteLLM
from ragbits.core.prompt import Prompt
from ragbits.core.vector_stores import InMemoryVectorStore
from ragbits.document_search import DocumentSearch
class QuestionAnswerPromptInput(BaseModel):
question: str
context: list[str]
class QuestionAnswerPromptOutput(BaseModel):
answer: str
class QuestionAnswerPrompt(Prompt[QuestionAnswerPromptInput, QuestionAnswerPromptOutput]):
system_prompt = """
You are a question answering agent. Answer the question that will be provided using context.
If in the given context there is not enough information refuse to answer.
"""
user_prompt = """
Question: {{ question }}
Context: {% for item in context %}
{{ item }}
{%- endfor %}
"""
embedder = LiteLLMEmbedder(model_name="text-embedding-3-small")
vector_store = InMemoryVectorStore(embedder=embedder)
document_search = DocumentSearch(vector_store=vector_store)
llm = LiteLLM(model_name="gpt-4.1-nano", use_structured_output=True)
async def run() -> None:
question = "What are the key findings presented in this paper?"
await document_search.ingest("web://https://arxiv.org/pdf/1706.03762")
result = await document_search.search(question)
prompt = QuestionAnswerPrompt(QuestionAnswerPromptInput(
question=question,
context=[element.text_representation for element in result],
))
response = await llm.generate(prompt)
print(response.answer)
if __name__ == "__main__":
asyncio.run(run())To expose your RAG application through Ragbits UI:
from collections.abc import AsyncGenerator
from pydantic import BaseModel
from ragbits.chat.api import RagbitsAPI
from ragbits.chat.interface import ChatInterface
from ragbits.chat.interface.types import ChatContext, ChatResponse
from ragbits.core.embeddings import LiteLLMEmbedder
from ragbits.core.llms import LiteLLM
from ragbits.core.prompt import Prompt
from ragbits.core.prompt.base import ChatFormat
from ragbits.core.vector_stores import InMemoryVectorStore
from ragbits.document_search import DocumentSearch
class QuestionAnswerPromptInput(BaseModel):
question: str
context: list[str]
class QuestionAnswerPrompt(Prompt[QuestionAnswerPromptInput, str]):
system_prompt = """
You are a question answering agent. Answer the question that will be provided using context.
If in the given context there is not enough information refuse to answer.
"""
user_prompt = """
Question: {{ question }}
Context: {% for item in context %}{{ item }}{%- endfor %}
"""
class MyChat(ChatInterface):
"""Chat interface for fullapp application."""
async def setup(self) -> None:
self.embedder = LiteLLMEmbedder(model_name="text-embedding-3-small")
self.vector_store = InMemoryVectorStore(embedder=self.embedder)
self.document_search = DocumentSearch(vector_store=self.vector_store)
self.llm = LiteLLM(model_name="gpt-4.1-nano", use_structured_output=True)
await self.document_search.ingest("web://https://arxiv.org/pdf/1706.03762")
async def chat(
self,
message: str,
history: ChatFormat | None = None,
context: ChatContext | None = None,
) -> AsyncGenerator[ChatResponse, None]:
# Search for relevant documents
result = await self.document_search.search(message)
prompt = QuestionAnswerPrompt(
QuestionAnswerPromptInput(
question=message,
context=[element.text_representation for element in result],
)
)
# Stream the response from the LLM
async for chunk in self.llm.generate_streaming(prompt):
yield self.create_text_response(chunk)
if __name__ == "__main__":
RagbitsAPI(MyChat).run()Create Ragbits projects from templates:
uvx create-ragbits-appExplore create-ragbits-app repo here. If you have a new idea for a template, feel free to contribute!
- Quickstart - Get started with Ragbits in a few minutes
- How-to - Learn how to use Ragbits in your projects
- CLI - Learn how to run Ragbits in your terminal
- API reference - Explore the underlying Ragbits API
We welcome contributions! Please read CONTRIBUTING.md for more information.
Ragbits is licensed under the MIT License.
