Build a Retrieval Augmented Generation (RAG) Agent

Using Julep, you can create a RAG Agent without having to play around with vector databases like ChromaDB, Weaviate etc.

Julep offers a pre-built RAG pipeline out of the box. You can specify data sources scoped to an agent or a user.

Preview

We'll build an Agent in <50 lines of code that has context over a blog post: LLM Powered Autonomous Agents by Lilian Weng.

from julep import Client
import os, bs4
from langchain_community.document_loaders import WebBaseLoader
from langchain_text_splitters import RecursiveCharacterTextSplitter

api_key = os.environ['API_KEY']
base_url = os.environ['BASE_URL']

client = Client(api_key=api_key, base_url=base_url)

agent = client.agents.create(
    name="Archy",
    about="A self-aware agent who knows a lot about LLMs, autonomous and agentic apps.",
    model="gpt-4o",
    metadata={"name": "Archy"}
)

docs = WebBaseLoader(
    web_paths=("https://lilianweng.github.io/posts/2023-06-23-agent/",),
    bs_kwargs=dict(parse_only=bs4.SoupStrainer(class_=("post-content", "post-title", "post-header")))
).load()

splits = RecursiveCharacterTextSplitter(chunk_size=2000, chunk_overlap=100).split_documents(docs)

for i, split in enumerate(splits):
    client.docs.create(
        agent_id=agent.id,
        doc={
            "title": "LLM Powered Autonomous Agents",
            "content": split.page_content,
            "metadata": {"chunk": i, **split.metadata},
        }
    )

session = client.sessions.create(
    agent_id=agent.id,
    situation="You are Ahti. You know all about AI and Agents",
    metadata={"agent_id": agent.id}
)

response = client.sessions.chat(
    session_id=session.id,
    messages=[{"role": "user", "content": "What are autonomous agents?"}],
    max_tokens=4096
)

(response_content,), doc_ids = response.response[0], response.doc_ids
print(f"{response_content.content}\n\nDocs used: {doc_ids}")

Installation & Set-Up

from julep import Client
from dotenv import load_dotenv
import os

load_dotenv()

client = Client(api_key=api_key, base_url=base_url)

Creating an Agent

agent = client.agents.create(
    name="Archy",
    about="A self aware agent who knows a lot about LLMs, autonomous and agentic apps.",
    model="gpt-4o",
    metadata={"name": "Ahti"},
)

Chunking

You can also use your chunking strategy. We recommend playing with the chunking strategy for best results over your use case.

Different chunk sizes, strategies have varied results over the accuracy of your RAG Agent!

## Downloading and splitting a webpage to add to docs
from langchain_community.document_loaders import WebBaseLoader
from langchain_text_splitters import RecursiveCharacterTextSplitter
import bs4

loader = WebBaseLoader(
    web_paths=("https://lilianweng.github.io/posts/2023-06-23-agent/",),
    bs_kwargs=dict(
        parse_only=bs4.SoupStrainer(
            class_=("post-content", "post-title", "post-header")
        )
    ),
)
docs = loader.load()
text_splitter = RecursiveCharacterTextSplitter(chunk_size=2000, chunk_overlap=100)
splits = text_splitter.split_documents(docs)

Indexing/Adding Chunks

Here the document is scoped to the agent, so it's the agent that has the context about this blog.

It is also possible to scope a document to a user. This is particularly useful when there are different users and there's a document that needs to be added for each of them.

for chunk_number, split in enumerate(splits):
    # Create a doc with the chunked content
    client.docs.create(
        agent_id=agent.id,
        doc={
            "title": "LLM Powered Autonomous Agents",
            "content": split.page_content,
            "metadata": {"chunk": chunk_number, **split.metadata},
        },
    )

Create Session

session = client.sessions.create(
    agent_id=agent.id,
    situation="You are Ahti. You know all about AI and Agents",
    metadata={"agent_id": agent.id},
)

Invoke Agent

The response object also returns which doc_ids were used in generating a message.

user_msg = "What are autonomous agents?"
response = client.sessions.chat(
    session_id=session.id,
    messages=[
        {
            "role": "user",
            "content": user_msg,
        }
    ],
    max_tokens=4096,
    recall=True,
    remember=True,
)

# refer to doc_ids in the response
print(f"{response.response[0][0].content}\n\n Docs used:{response.doc_ids}")