How to integrate SIE with your existing stack

SIE provides three integration paths depending on how you are building. Framework adapters plug SIE into existing RAG pipelines with minimal code. The native SDK gives you full access to every feature. OpenAI compatibility lets you point existing OpenAI code at SIE with a single URL change.

Option	Best for
Framework adapters	LangChain, LlamaIndex, Haystack, Qdrant, Weaviate, Chroma, LanceDB, DSPy, CrewAI
Native SDK	Custom pipelines, multi-vector output, full extraction support
OpenAI compatibility	Migrating existing OpenAI embeddings or chat code

Which Integration Path Should I Use?

Use the summary table and anchor links above to jump to framework adapters, the native SDK, or OpenAI compatibility.

Framework Adapters

SIE provides dedicated packages for all major Python AI frameworks and vector stores, plus four TypeScript packages.

Python

Framework	Package	Embeddings	Sparse	Multivector	Reranking	Extraction	Multimodal
Chroma	`sie-chroma`	Yes	Yes	No	No	No	No
CrewAI	`sie-crewai`	No	Yes	No	Yes	Yes	No
DSPy	`sie-dspy`	Yes	Yes	No	Yes	Yes	No
Haystack	`sie-haystack`	Yes	Yes	Yes	Yes	Yes	Yes
LanceDB	`sie-lancedb`	Yes	No	No	Yes	Yes	No
LangChain	`sie-langchain`	Yes	Yes	No	Yes	Yes	No
LlamaIndex	`sie-llamaindex`	Yes	Yes	No	Yes	Yes	Yes
Qdrant	`sie-qdrant`	Yes	Yes	Yes	No	No	No
Weaviate	`sie-weaviate`	Yes	Yes	Yes	No	No	No

TypeScript

Framework	Package	Embeddings	Sparse	Multivector	Reranking	Extraction
Chroma	`@superlinked/sie-chroma`	Yes	Yes	No	No	No
LanceDB	`@superlinked/sie-lancedb`	Yes	No	No	Yes	No
LangChain.js	`@superlinked/sie-langchain`	Yes	Yes	No	Yes	Yes
LlamaIndex.ts	`@superlinked/sie-llamaindex`	Yes	Yes	No	Yes	Yes

Use a framework adapter when:

You are building a RAG pipeline with one of these frameworks
You need sparse embeddings for hybrid search
You need reranking to improve retrieval quality

For features marked “No” in the table above, you can use the native SDK alongside your framework integration. See each integration’s page for examples.

Native SDK

The native SDK gives you full access to all SIE features: dense, sparse, and multi-vector embeddings, reranking, and extraction (entities, relations, classifications, object detection).

pip install sie-sdk
# or
pnpm add @superlinked/sie-sdk

Python
TypeScript

from sie_sdk import SIEClient
from sie_sdk.types import Item

client = SIEClient("http://localhost:8080")

# All output types from one model
result = client.encode(
    "BAAI/bge-m3",
    Item(text="Your text"),
    output_types=["dense", "sparse", "multivector"]
)

# Reranking
scores = client.score(
    "BAAI/bge-reranker-v2-m3",
    query=Item(text="What is AI?"),
    items=[Item(text="AI is..."), Item(text="Weather is...")]
)

# Entity extraction
entities = client.extract(
    "urchade/gliner_multi-v2.1",
    Item(text="Tim Cook leads Apple."),
    labels=["person", "organization"]
)

import { SIEClient } from "@superlinked/sie-sdk";

const client = new SIEClient("http://localhost:8080");

// All output types from one model
const result = await client.encode(
  "BAAI/bge-m3",
  { text: "Your text" },
  { outputTypes: ["dense", "sparse", "multivector"] },
);

// Reranking
const scores = await client.score(
  "BAAI/bge-reranker-v2-m3",
  { text: "What is AI?" },
  [{ text: "AI is..." }, { text: "Weather is..." }],
);

// Entity extraction
const entities = await client.extract(
  "urchade/gliner_multi-v2.1",
  { text: "Tim Cook leads Apple." },
  { labels: ["person", "organization"] },
);

await client.close();

Use the native SDK when:

You are building a custom pipeline without a framework
You need multi-vector (ColBERT) output
You need extraction (entities, relations, classifications, object detection)
You want fine-grained control over batching and timing

OpenAI Compatibility

SIE speaks the OpenAI API for both embeddings and generation. /v1/embeddings and /v1/chat/completions are drop-in: existing OpenAI code works with a single URL change and no other modifications. The gateway also serves /v1/completions (non-streaming) and /v1/responses (string input, non-streaming) for the common cases:

from openai import OpenAI

client = OpenAI(base_url="http://localhost:8080/v1", api_key="not-needed")

response = client.embeddings.create(
    model="BAAI/bge-m3",
    input=["Your text here", "Another text"]
)
for item in response.data:
    print(f"Index {item.index}: {len(item.embedding)} dimensions")

reply = client.chat.completions.create(
    model="Qwen/Qwen3-4B-Instruct-2507",
    messages=[{"role": "user", "content": "Reply with one word: the capital of France."}],
)
print(reply.choices[0].message.content)

Where each endpoint is served: /v1/embeddings is available on every SIE server, single-node or cluster. /v1/chat/completions, /v1/completions, and /v1/responses are served by the cluster gateway against generation models (the sglang bundle). On a single-node Apple Silicon server, /v1/chat/completions also works against the local MLX generation backend.

Use OpenAI compatibility when:

You have existing code using the OpenAI SDK
You need dense embeddings or chat completions
You want zero code changes beyond the URL

Limitations: Sparse, multi-vector, reranking, and extraction are not part of the OpenAI API shape; they require the native SDK or a framework adapter.

Full Feature Comparison

Feature	Framework adapters	Native SDK	OpenAI compat
Dense embeddings	All	Yes	Yes
Chat / text generation	No	Yes	Yes (gateway)
Sparse embeddings	Most	Yes	No
Multi-vector (ColBERT)	Haystack, Qdrant, Weaviate	Yes	No
Reranking	Haystack, LangChain, LlamaIndex, LanceDB	Yes	No
Extraction (NER, relations, classification, vision)	Haystack, LangChain, LlamaIndex, LanceDB, CrewAI, DSPy	Yes	No

Frequently Asked Questions

Does SIE work with Qdrant? Yes. Install sie-qdrant for dense and sparse embedding support with Qdrant. See the Qdrant integration guide.

Can I use SIE as a drop-in replacement for OpenAI? Yes, for dense embeddings and chat completions. Change your base URL to your SIE server and set any string as the API key. The response format matches OpenAI’s exactly. The gateway also serves /v1/completions and /v1/responses, currently non-streaming. See OpenAI compatibility above.

Which integration supports the most features? sie-haystack and sie-llamaindex support the most features, including embeddings, sparse, reranking, extraction, and multimodal. See the full feature table above.

Where can I find setup instructions for each framework? Each framework has its own dedicated page. See Chroma, LangChain, LlamaIndex, Haystack, Qdrant, Weaviate, LanceDB, DSPy, and CrewAI.