LlamaIndex Networks

The llama-index-networks library extension allows for the creation of networks knowledge that can be queried and used for LLM context augmentation. In such networks, we have "data suppliers" and "data consumers" that participate and benefit from the network, respectively.

Data suppliers

With this extension, you can easily wrap your llama-index QueryEngine with a ContributorService that exposes the query engine behind a REST (FastAPI) service, which renders it ready to contribute to LlamaIndex Network!

from llama_index.networks.contributor import (
    ContributorServiceSettings,
    ContributorService,
)

query_engine = ...  # build a query engine as typically done with llama-index

settings = ContributorServiceSettings()
service = ContributorService(config=settings, query_engine=query_engine)
app = service.app


if __name__ == "__main__":
    import uvicorn

    uvicorn.run(app, port=8000)

Data consumers

With llama-index-networks, you can build a NetworkQueryEngine that is able to connect to set of (network) ContributorService's.

from llama_index.networks.contributor import ContributorClient
from llama_index.networks.query_engine import NetworkQueryEngine
from llama_index.llms.groq import Groq

# Use ContributorClient to connect to a ContributorService
client = ContributorClient.from_config_file(env_file=".env.contributor.client")
contributors = [client]

# NetworkQueryEngine
llm = Groq()
network_query_engine = NetworkQueryEngine.from_args(
    contributors=contributors, llm=llm
)

# Can query it like any other query engine
network_query_engine.query("Why is the sky blue?")

The .env.contributor.client file contains the parameters to connect to the ContributorService (i.e., api_url and api_token.)

Examples

For a full demo, checkout the examples/demo sub-folder.

FAQ

1. Can I add my own custom endpoints to a ContributorService?

Yes, the (Fastapi) app is accessible via ContributorService.app and can be modified to your needs.

settings = ContributorServiceSettings()
service = ContributorService(config=settings, query_engine=query_engine)
app = service.app


@app.get("<custom-path>")
def custom_endpoint_logic():
    ...

2. How can I add authentication to a ContributorService?

Currently, the client supports authentication through request headers as well as as via an api_token str. On the server side, you should be able to modify the app to your needs:

settings = ContributorServiceSettings()
service = ContributorService(config=settings, query_engine=query_engine)
app = service.app

# modify app here to include your own security mechanisms

As we continue to build out this library extension, we welcome any feedback or suggestions on what can be incorporated on this regard and others in order to make our tools more helpful to our users.

📖 Citation

Reference to cite if you use LlamaIndex Networks in a paper, please cite both the main library as well as this extension.

@software{Liu_LlamaIndex_2022,
author = {Liu, Jerry},
doi = {10.5281/zenodo.1234},
month = {11},
title = {{LlamaIndex}},
url = {https://github.com/jerryjliu/llama_index},
year = {2022}
}

@software{Fajardo_LlamaIndexNetworks_2024,
author = {Fajardo, Val Andrei and Liu, Jerry and Markewich, Logan and Suo, Simon and Zhang, Haotian and Desai, Sourabh},
doi = {10.5281/zenodo.1234},
month = {11},
title = {{LlamaIndex Networks}},
url = {https://github.com/run-llama/llama_index/llama-index-networks},
year = {2024}
}