Skip to content

Configuration

General

The configuration of the ERI-Server can be done using the eri.ini configuration file.

This file is split into multiple sections:

  • A general configuration of the ERI-Server. This section modifies the behaviour of the entire interface.
  • An optional list of sections for the provided retrieval algorithm. All retrieval algorithms configured within such a section are available for the user when querying the interface

Configuring the Interface

To configure the general interface settings, found in the ERI Settings section, please follow these steps closely:

  1. Label your data source: To enable users to easily identify what information can be retrieved using the interface, a name and description must be provided. These can be entered into the data_source_name and data_source_description fields.
  2. Set privacy settings: To restrict the usage of external LLMs on internal data, the type of allowed LLM providers can be set. Please enter one of SELF_HOSTED | ANY into the allowed_provider_type field.
  3. Configure data access: To limit the access to the ERI-Server, and thus to the information handled by such, the type of available authorization method can be set. Please enter one of NONE | TOKEN into the authorization_methods fields or, to enable multiple options, enter a JSON formatted list of the respective methods as strings instead. For more information about the available authorization methods, please refer to authorization methods.
  4. [Optional] Enable https: To start the Server on HTTPS, please provide the paths to a valid SSL certificate and key in the ssl_cert_path and ssl_key_path fields, respectively. The Server is started in HTTPS mode if both fields are provided; otherwise, HTTP is used instead.
  5. Setup the authentication database Not necessary when only using NONE authentication: The needed data for authentication validation is stored in a postgres database. The connection to this database has to be specified here. Please provide a JSON formatted dictionary with the following properties to the postgres_connection option: username, password, url, database_name. For more information about the database, refer to authorization methods.

To enable selected retrieval algorithms, follow the steps specific to the desired method.

GARAG

To enable GARAG as a retrieval process, please paste the following skeleton into the configuration file and follow these steps closely:

[GARAG]
config = 
emb_model = 
elastic_db_url = 
arango_db =
  1. Enable custom configuration: To configure the retriever, please provide a JSON formatted dictionary using the properties below in the config field.
    • top_k (int; 0 < x): The maximum number of hits to retrieve. May be set by an user during a retrieval request.
    • confidence_cutoff (float; 0 < x < 1): A value to specify a minimum similiarity between the query and the retrieved information. May be set by an user during a retrieval request.
    • vector_db_index_name (string): The index to be used for searching relevant information.
  2. Specify the embedding model used: Enter the name of the embedding model used to convert requests into vectors into the emb_model field. This has to match the model used during the indexing phase.
  3. Set the Elasticsearch connection: Enter the url to the Elasticsearch server into the elastic_db_url field, on which the vector index can be found.
  4. Set the ArangoDB connection: Please specify the connection to ArangoDB as a JSON formatted dictionary using the properties below in the arango_db field. The graph_name should point to the community graph build during the indexing phase.

GARAG config

{
    "top_k": int,
    "confidence_cutoff": float,
    "vector_db_index_name": string
}

ArangoDB config

{
    "url": string,
    "username": string,
    "password": string,
    "db_name": string,
    "graph_name": string
}

GraphRAG

To enable GraphRAG as a retrieval process, please paste the following skeleton into the configuration file and follow these steps closely:

[GraphRAG]
config = 
llm = 
arango_db =
  1. Enable custom configuration: To configure the retriever, please provide a JSON formatted dictionary using the properties below in the config field.
    • top_k (int; 0 < x): The maximum number of hits to retrieve. May be set by an user during a retrieval request.
    • confidence_cutoff (int; 0 < x < 100): A value to specify a minimum similiarity between the query and the retrieved information. May be set by an user during a retrieval request.
    • community_degree (int; 0 < x): The maximum community degree to be used during retrieval. A higher value will incorporate more precise summaries, resulting in a higher node count and longer query time. May be set by an user during a retrieval request.
  2. Set the LLM connection: Please specify the connection to a LLM service as a JSON formatted dictionary using the properties below in the llm field. This LLM is used to gernerate and evaluate partial answers.
  3. Set the ArangoDB connection: Please specify the connection to ArangoDB as a JSON formatted dictionary using the properties below in the arango_db field. The graph_name should point to the community graph build during the indexing phase.

GraphRAG config

{
    "top_k": int,
    "confidence_cutoff": int,
    "community_degree": int
}

LLM config

{
    "provider": "ollama" | "openai",
    "base_url": string,
    "api_key": string (optional),
    "model_name": string (optional),
    "options": dictionary
}

ArangoDB config

{
    "url": string,
    "username": string,
    "password": string,
    "db_name": string,
    "graph_name": string
}

NaiveGraphRAG

To enable NaiveGraphRAG as a retrieval process, please paste the following skeleton into the configuration file and follow these steps closely:

[NaiveGraphRAG]
config = 
emb_model = 
elastic_db_url =
  1. Enable custom configuration: To configure the retriever, please provide a JSON formatted dictionary using the properties below in the config field.
    • top_k (int; 0 < x): The maximum number of hits to retrieve. May be set by an user during a retrieval request.
    • confidence_cutoff (float; 0 < x < 1): A value to specify a minimum similiarity between the query and the retrieved information. May be set by an user during a retrieval request.
    • vector_db_index_name (string): The index to be used for searching relevant information.
  2. Specify the embedding model used: Enter the name of the embedding model used to convert requests into vectors into the emb_model field. This has to match the model used during the indexing phase.
  3. Set the Elasticsearch connection: Enter the url to the Elasticsearch server into the elastic_db_url field, on which the vector index can be found.

NaiveGraphRAG config

{
    "top_k": int,
    "confidence_cutoff": float,
    "vector_db_index_name": string
}

NaiveRAG

To enable NaiveRAG as a retrieval process, please paste the following skeleton into the configuration file and follow these steps closely:

[NaiveRAG]
config = 
emb_model = 
elastic_db_url =
  1. Enable custom configuration: To configure the retriever, please provide a JSON formatted dictionary using the properties below in the config field.
    • top_k (int; 0 < x): The maximum number of hits to retrieve. May be set by an user during a retrieval request.
    • confidence_cutoff (float; 0 < x < 1): A value to specify a minimum similiarity between the query and the retrieved information. May be set by an user during a retrieval request.
    • vector_db_index_name (string): The index to be used for searching relevant information.
  2. Specify the embedding model used: Enter the name of the embedding model used to convert requests into vectors into the emb_model field. This has to match the model used during the indexing phase.
  3. Set the Elasticsearch connection: Enter the url to the Elasticsearch server into the elastic_db_url field, on which the vector index can be found.

NaiveRAG config

{
    "top_k": int,
    "confidence_cutoff": float,
    "vector_db_index_name": string
}

VectorGR

To enable VectorGR as a retrieval process, please paste the following skeleton into the configuration file and follow these steps closely:

[VectorGR]
config = 
emb_model = 
llm = 
neo4j_db =
  1. Enable custom configuration: To configure the retriever, please provide a JSON formatted dictionary using the properties below in the config field.
    • top_k (int; 0 < x): The maximum number of hits to retrieve. May be set by an user during a retrieval request.
    • v_index_name (string):
    • return_properties (List[string]): List of node properties to return.
    • filters (Dictionary[string, Any]): Filters for metadata pre-filtering. When performing a similarity search, one may have constraints to apply. May be set by an user during a retrieval request.
  2. Specify the embedding model used: Enter the name of the embedding model used to convert requests into vectors into the emb_model field. This has to match the model used during the indexing phase.
  3. Set the LLM connection: Please specify the connection to a LLM service as a JSON formatted dictionary using the properties below in the llm field. This LLM is used to gernerate and evaluate partial answers.
  4. Set the Neo4J connection: Please specify the connection to ArangoDB as a JSON formatted dictionary using the properties below in the neo4j_db field. The graph_name should point to the community graph build during the indexing phase.

VectorGR config

{
    "top_k": int,
    "v_index_name": string,
    "return_properties": List[string],
    "filters": Dictionary[string, Any]
}

LLM config

{
    "provider": "ollama" | "openai",
    "base_url": string,
    "api_key": string (optional),
    "model_name": string (optional),
    "options": dictionary
}

Neo4JDB config

{
    "url": string,
    "db_name": string,
    "password": string
}

HybridGR

To enable HybridGR as a retrieval process, please paste the following skeleton into the configuration file and follow these steps closely:

[HybridGR]
config = 
emb_model = 
llm = 
neo4j_db =
  1. Enable custom configuration: To configure the retriever, please provide a JSON formatted dictionary using the properties below in the config field.
    • top_k (int; 0 < x): The maximum number of hits to retrieve. May be set by an user during a retrieval request.
    • v_index_name (string):
    • f_index_name (string):
    • return_properties (List[string]): List of node properties to return.
  2. Specify the embedding model used: Enter the name of the embedding model used to convert requests into vectors into the emb_model field. This has to match the model used during the indexing phase.
  3. Set the LLM connection: Please specify the connection to a LLM service as a JSON formatted dictionary using the properties below in the llm field. This LLM is used to gernerate and evaluate partial answers.
  4. Set the Neo4J connection: Please specify the connection to ArangoDB as a JSON formatted dictionary using the properties below in the neo4j_db field. The graph_name should point to the community graph build during the indexing phase.

HybridGR config

{
    "top_k": int,
    "v_index_name": string,
    "f_index_name": string,
    "return_properties": List[string]
}

LLM config

{
    "provider": "ollama" | "openai",
    "base_url": string,
    "api_key": string (optional),
    "model_name": string (optional),
    "options": dictionary
}

Neo4JDB config

{
    "url": string,
    "db_name": string,
    "password": string
}

Text2Cypher

To enable Text2Cypher as a retrieval process, please paste the following skeleton into the configuration file and follow these steps closely:

[Text2Cypher]
config = 
llm = 
neo4j_db =
  1. Enable custom configuration: To configure the retriever, please provide a JSON formatted dictionary using the properties below in the config field.
    • examples: Optional user input/query pairs for the LLM to use as examples.
  2. Set the LLM connection: Please specify the connection to a LLM service as a JSON formatted dictionary using the properties below in the llm field. This LLM is used to gernerate and evaluate partial answers.
  3. Set the Neo4J connection: Please specify the connection to ArangoDB as a JSON formatted dictionary using the properties below in the neo4j_db field. The graph_name should point to the community graph build during the indexing phase.

Text2Cypher config

{
    "examples": List[string]
}

LLM config

{
    "provider": "ollama" | "openai",
    "base_url": string,
    "api_key": string (optional),
    "model_name": string (optional),
    "options": dictionary
}

Neo4JDB config

{
    "url": string,
    "db_name": string,
    "password": string
}

Authorization methods

NONE

This method is used to allow free access to all users. Everyone can generate a session token without restrictions, which can the be used to retrieve data from the data source.

TOKEN

This method retricts access to the data source by requireing an API key, provided as a Bearer token, when generating a session token. Valid API keys are stored in a Postgres database, allowing other services to manage these keys.

To use the TOKEN authorization method, a table named API_Key must be created in the Postgres database. This can be done using the following command: 

CREATE TABLE IF NOT EXISTS API_Key (
    key varchar PRIMARY KEY, 
    user varchar
);

This table is created automatialy on startup, if it does not already exist. During authentication, the provided token is compared to all key fields within the API_Key table. If a match is found, access is granted; otherwise, unauthorized access is denied.

Specifying the postgres connection

When a database is needed, please provide a JSON formatted dictionary using the following properties in the postgres_connection field in the ERI Settings section:

{
    "username": string,
    "password": string,
    "url": string,
    "database_name": string
}