For the knowledge graph created, which service is it stored in?

[chakpongchung]

I don't seem to see a graph store used in the project. For the topology part of the knowledge graph, which service is it stored in so people can query against?

[jgbradley1]

The knowledge graph is saved off as a collection of parquet files locally.

The output of an indexing run is defined by the storage section of the configuration file. See here for a description of all configuration values. For example, if storage.type=file, a directory with the name storage.base_dir (default value=output) will contain all the parquet files from the indexing run.

Other storage options are available (i.e. Azure Storage) which you can configure graphrag to use if you wish.

[hclgit2rahul]

The knowledge graph is saved off as a collection of parquet files locally.

The output of an indexing run is defined by the storage section of the configuration file. See here for a description of all configuration values. For example, if storage.type=file, a directory with the name storage.base_dir (default value=output) will contain all the parquet files from the indexing run.

Other storage options are available (i.e. Azure Storage) which you can configure graphrag to use if you wish.

@jgbradley1 : Thanks for clarifying the query on storage location. I have a follow-up question, based on mentioned configuration, I am able to store parquet files on azure blob location with similar folder structure as it was in local. However, the same configuration is not picked for graphrag.query, it continue to looks for local parquet files. Is there a switch or configuration which instruct it look it in blob storage ?

[jesusfbes]

I have the same question of @hclgit2rahul. @jgbradley1 would you mind to help us with this?

[hclgit2rahul]

@jesusfbes @jgbradley1 - I spent some more time, it seems that programmatically we can read data frames from paraquet file stored in azure blob and the use them for search and LLM prompts to get the outcome. However, using CLI - I do not see an option.

[AR-javis]

Is it possible to use it with AWS S3? Or is Azure the only one supported? Thanks in advance

[jgbradley1]

@hclgit2rahul @jesusfbes you are correct that the query CLI only worked on local parquet files. We have made several major changes internal to the package over the last several months to get indexing and query more in sync with each other and provide support for the same functionality (i.e. utilizing azure storage as a storage backend).

If you haven't tried the latest release of graphrag, please give it a try and open an issue if the problem still exist (I recall this particular issue did get addressed though).

[reachrkr]

How can we store in other formats used by popular Graph DB's like neo4j

[jgbradley1]

The quickest approach would be to write a small script that transforms the parquet file data to a format (i.e. CSV) that is acceptable by the graph DB that you wish to use. After running graphrag indexing, you would then want to run this data transformation script over the parquet files.

[jli113]

Already have the graph file and loaded it in neo4j, can we use graphs in GraphDB directly instead of the indexing process?

[esbenclarlund]

How to use the vector db in Azure. Preferable I would like to configure graphRAG to write to Azure ai search directly but I do not find that option but to get it into a blob will do.
I already got it to write the .parquet files to a blob on generation but do not know how to utilize what by default is written to lancedb on my local disk.
Any input will be highly appreciated, thanks.