LlamaIndex Cheatsheet: Documents, Retrievers, Agents and RAG

# Core package + the integrations you'll need
pip install llama-index                              # umbrella metapackage
pip install llama-index-core                         # if you want to pick & mix
pip install llama-index-llms-openai \
            llama-index-embeddings-openai \
            llama-index-readers-file \
            llama-index-vector-stores-chroma

# env
export OPENAI_API_KEY=sk-...
export ANTHROPIC_API_KEY=sk-...
export LLAMA_CLOUD_API_KEY=llx-...   # optional, LlamaParse / LlamaCloud

# Global defaults (used by every index/engine unless overridden)
python - <<'PY'
from llama_index.core import Settings
from llama_index.llms.openai import OpenAI
from llama_index.embeddings.openai import OpenAIEmbedding
Settings.llm = OpenAI(model="gpt-4o-mini", temperature=0)
Settings.embed_model = OpenAIEmbedding(model="text-embedding-3-small")
PY

from llama_index.core import VectorStoreIndex, StorageContext
from llama_index.core.postprocessor import SimilarityPostprocessor
from llama_index.core.response_synthesizers import get_response_synthesizer

# Reload the existing store as an index (no re-embedding)
index = VectorStoreIndex.from_vector_store(store)

# Hand-roll the pipeline: retriever -> postprocessors -> synthesizer
retriever = index.as_retriever(similarity_top_k=6)
postprocessors = [SimilarityPostprocessor(similarity_cutoff=0.7)]
synth = get_response_synthesizer(response_mode="compact")

# Or use the prebuilt query engine
query_engine = index.as_query_engine(
    similarity_top_k=6,
    response_mode="compact",
    node_postprocessors=postprocessors,
)

resp = query_engine.query("What does the onboarding doc say about access?")
print(str(resp))
for n in resp.source_nodes:
    print(n.score, n.node.metadata.get("file_name"))

from llama_index.core import SimpleDirectoryReader
from llama_index.core.node_parser import SentenceSplitter
from llama_index.core.ingestion import IngestionPipeline
from llama_index.embeddings.openai import OpenAIEmbedding
from llama_index.vector_stores.chroma import ChromaVectorStore
import chromadb

# 1 · Load
docs = SimpleDirectoryReader("./data", recursive=True).load_data()

# 2 · Build a reusable pipeline (splitter + embeddings + sink)
chroma = chromadb.PersistentClient(path="./chroma")
store = ChromaVectorStore(chroma_collection=chroma.get_or_create_collection("docs"))

pipeline = IngestionPipeline(
    transformations=[
        SentenceSplitter(chunk_size=800, chunk_overlap=120),
        OpenAIEmbedding(model="text-embedding-3-small"),
    ],
    vector_store=store,
)

# 3 · Run — produces nodes and writes them to the vector store
nodes = pipeline.run(documents=docs, show_progress=True)

from llama_index.core.tools import FunctionTool, QueryEngineTool
from llama_index.core.agent.workflow import FunctionAgent
from llama_index.llms.openai import OpenAI

def add(a: int, b: int) -> int:
    """Return a + b."""
    return a + b

calc = FunctionTool.from_defaults(fn=add)

docs_tool = QueryEngineTool.from_defaults(
    query_engine=query_engine,
    name="company_docs",
    description="Search internal onboarding + HR docs.",
)

agent = FunctionAgent(
    tools=[calc, docs_tool],
    llm=OpenAI(model="gpt-4o-mini"),
    system_prompt="You are a precise assistant. Cite sources.",
)

resp = await agent.run("How many vacation days do new hires get? Then add 3.")
print(resp)

from llama_index.core import (
    SimpleDirectoryReader, VectorStoreIndex, Settings, StorageContext,
)
from llama_index.core.node_parser import SentenceSplitter
from llama_index.llms.openai import OpenAI
from llama_index.embeddings.openai import OpenAIEmbedding
from llama_index.vector_stores.chroma import ChromaVectorStore
import chromadb

# 1 · Configure global defaults
Settings.llm = OpenAI(model="gpt-4o-mini", temperature=0)
Settings.embed_model = OpenAIEmbedding(model="text-embedding-3-small")
Settings.node_parser = SentenceSplitter(chunk_size=800, chunk_overlap=120)

# 2 · Persistent vector store
client = chromadb.PersistentClient(path="./chroma")
store = ChromaVectorStore(chroma_collection=client.get_or_create_collection("docs"))
storage = StorageContext.from_defaults(vector_store=store)

# 3 · Build (or rebuild) the index from a folder
docs = SimpleDirectoryReader("./data", recursive=True).load_data()
index = VectorStoreIndex.from_documents(docs, storage_context=storage)

# 4 · Ask
engine = index.as_query_engine(similarity_top_k=4, response_mode="compact")
print(engine.query("Summarise the access policy in 3 bullets."))