This tool combines the Neo4j Vector search capabilities with embeddings from OpenAI.
The code itself is working for the example below, but it is in a WIP state.
You can create your own binaries by invoking ./mvnw package.
For creating native images, it is required to run the build process with a Graal JVM.
The most common workflow would be to create embeddings first and then query those embeddings with search phrases.
You have to have an OpenAI API key to use this tool.
The tool expects the OPENAI_API_KEY environment variable to be set with a working key.
Note
Please note that the --label parameter has to be set on the main command before invoking the subcommand.
Example usage
> neo4j-vector-cli --label Game create-embedding --properties name,about,descriptionThis command will create embeddings for all nodes labeled :Game and join its fields name, about, and description
to be tokenized.
After getting the results back, it will store those on the nodes' property embedding per default.
This property is customizable, please have a look at the Configuration section.
Example usage
> neo4j-vector-cli --label Game search 'fun family strategy'
{records:[{__elementId__:"4:2..",__similarity__:0.8931642},{__elementId__:"4:1..",__similarity__:0.892818},...]}The default output format might surprise at the first glance, but this output can be taken directly as input for Neo4j's Cypher Shell.
Output as Cypher Shell input
> cypher-shell 'UNWIND $records as record MATCH (g:Game)-[:DEVELOPED_BY]-(d) WHERE elementId(g) = record.__elementId__ RETURN g.name as game_name, g.metacritic_score as metacritic_score, d.name as developer, record.__similarity__ as similarity'"
Type " -P $(neo4j-vector-cli --label Game search 'fun family strategy')This gives you the option to combine the vector search results with arbitrary Cypher statements for further processing.
If you are just interested in the results from the vector search themselves, use the --format console output.
> neo4j-vector-cli --label Game search 'family calm peaceful fun' --format console -p name
Labels Properties Similarity
[Game] [name=Sheltered] 0.88821214
[Game] [name=Alba: A Wildlife Adventure] 0.88753295
[Game] [name=Before We Leave] 0.88637364
[Game] [name=Shelter] 0.88603795
[Game] [name=Deep Sea Tycoon: Diver\'s Paradise] 0.88546705
You can create a property file named ~/.neo4j-vector-cli.properties containing the configuration:
neo4j-vector-cli.uri = neo4j://localhost:7687
neo4j-vector-cli.password = maybe_secureAll properties can be found by invoking the command neo4j-vector-cli:
Usage: neo4j-vector-cli [-v] [--embedding-property=<embeddingProperty>]
--label=<label> [--model=<model>]
[--password=<password>] [--uri=<uri>] [--user=<user>]
[COMMAND]
--embedding-property=<embeddingProperty>
Property to be used for storing and searching the
embedding.
Defaults to 'embedding'.
--label=<label> Node label to be used for the operation.
--model=<model> Training model to be used.Defaults to
'text-embedding-ada-002'.
--password=<password>
Neo4j password to be used.
--uri=<uri> Neo4j server URI to connect to.
Defaults to 'bolt://localhost:7687'.
--user=<user> Neo4j user to be used.
Defaults to 'neo4j'.
-v, --verbose Verbose logging, defaults to 'false'.
Commands:
search Search with any phrase in the embeddings.
create-embedding Create embeddingor its subcommands.
search
Usage: neo4j-vector-cli search [-f=<format>] [-l=<limit>] [-t=<threshold>]
[-p=property[,property...]]... <phrase>
Search with any phrase in the embeddings.
<phrase> The search phrase.
-f, --format=<format> Output format, can be either 'parameter' or 'console'.
Parameter output can be used e.g. as input for
cypher-shell's -p(arameter) option.
Defaults to 'parameter'.
-l, --limit=<limit> Result size for search and output.
-p, --properties=property[,property...]
Node properties to include in the result.
-t, --threshold=<threshold>
Cap of accepted lower bound similarity (0.0 - 1.0).create-embedding
Usage: neo4j-vector-cli create-embedding -p=property[,property...] [-p=property
[,property...]]...
Create embedding
-p, --properties=property[,property...]
Node properties to include during embedding creation.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent