LlamaIndex - Neosantara Docs

LlamaIndex is a data framework designed to connect private and enterprise data to large language models through Retrieval-Augmented Generation (RAG). Neosantara provides OpenAI-compatible endpoints for both text generation and vector embeddings.

Setup

Install the LlamaIndex packages and OpenAI-compatible adapters:

pip install -U llama-index llama-index-llms-openai-like llama-index-embeddings-openai-like

export NEOSANTARA_API_KEY="nsk_your_api_key_here"

NEOSANTARA_API_KEY=nsk_your_api_key_here

$env:NEOSANTARA_API_KEY="nsk_your_api_key_here"

Global Settings Configuration

Use the LlamaIndex Settings module to set both the generation LLM and the embedding provider globally.

import os
from llama_index.core import Settings
from llama_index.llms.openai_like import OpenAILike
from llama_index.embeddings.openai_like import OpenAILikeEmbedding

Settings.llm = OpenAILike(
    model="deepseek-v4.1-flash",
    api_key=os.environ["NEOSANTARA_API_KEY"],
    api_base="https://api.neosantara.xyz/v1",
    temperature=0.2,
    max_tokens=1024,
    is_chat_model=True
)

Settings.embed_model = OpenAILikeEmbedding(
    model="text-embedding-3-small",
    api_key=os.environ["NEOSANTARA_API_KEY"],
    api_base="https://api.neosantara.xyz/v1"
)

End-to-End RAG Pipeline Example

The following script creates an in-memory vector index from text documents and queries the index with synthesized context:

import os
from llama_index.core import Document, VectorStoreIndex, Settings
from llama_index.llms.openai_like import OpenAILike
from llama_index.embeddings.openai_like import OpenAILikeEmbedding

# 1. Configure generation and embedding models
Settings.llm = OpenAILike(
    model="deepseek-v4.1-flash",
    api_key=os.environ["NEOSANTARA_API_KEY"],
    api_base="https://api.neosantara.xyz/v1",
    is_chat_model=True
)

Settings.embed_model = OpenAILikeEmbedding(
    model="text-embedding-3-small",
    api_key=os.environ["NEOSANTARA_API_KEY"],
    api_base="https://api.neosantara.xyz/v1"
)

# 2. Define source documents
documents = [
    Document(text="Neosantara is Indonesia's AI Gateway providing Rupiah PAYG billing and multi-provider routing."),
    Document(text="The Free tier rate limits are 10 RPM, 30,000 ITPM, and 8,000 OTPM with a minimum deposit floor of IDR 50,000."),
    Document(text="The Indonesian UU PDP No. 27/2022 automated PII guardrail is enabled per-request via the X-Guard: on header.")
]

# 3. Build vector index
index = VectorStoreIndex.from_documents(documents)

# 4. Instantiate query engine and ask questions
query_engine = index.as_query_engine(similarity_top_k=2)
response = query_engine.query("How do developers enable Indonesian PII redaction on Neosantara?")

print("Answer:", response)

Streaming Responses in LlamaIndex

For conversational chat interfaces or interactive CLIs, enable streaming on the query engine:

query_engine = index.as_query_engine(streaming=True)
streaming_response = query_engine.query("What are the rate limits for the Free tier?")

for text in streaming_response.response_gen:
    print(text, end="", flush=True)
print()

Recommended RAG Model Pairings

Workflow	Embedding Model	Generation Model (LLM)
High Throughput & Low Cost	`text-embedding-3-small`	`gemini-3.8-flash`
Deep Synthesis & Long Documents	`text-embedding-3-small`	`deepseek-v4.1-flash`
High Precision & Legal Compliance	`text-embedding-3-large`	`claude-4.5-sonnet`

Next Steps

Related topics

Embeddings Data Guardrails & UU PDP LiteLLM Model Catalog

⌘I