config module provides an interface for reading and writing configuration values. It includes functions to load, retrieve, and update configuration values for the Prometheux platform.
Configuration Functions
Get Configuration Value
Retrieves a value from the configuration using the specified key.- Python SDK
- REST API
ReturnsThe value associated with the specified key, or
None if the key does not exist.Set Configuration Value
Sets a configuration value for the specified key.- Python SDK
- REST API
Update Multiple Configuration Values
Update multiple configuration values at once from a dictionary.- Python SDK
- REST API
Authentication Token
ThePMTX_TOKEN can be configured in two ways:
Using Environment Variables
Using the SDK Configuration
Configuring LLMs
The Prometheux Chain exploits LLMs to generate more human-readableexplanations, translations of Vadalog rules, graph rag and LLM output validation tasks. By default, no LLM is configured.
Setting up the LLM
To enable and configure an LLM, such as OpenAI’s GPT, follow these steps: 1. Set the LLM Provider:Default Configurations
The SDK uses the following default configurations:User Management
Get User Role
Retrieve the current user’s role.- Python SDK
- REST API
Get Usage Status
Get current API usage statistics and limits.- Python SDK
- REST API
Get Login Activity
Return the distinct IP addresses that have accessed your workspace, with geolocation and access timestamps, most recent first.- Python SDK
- REST API
List LLM Models
Return the available model names for a given LLM provider. The request body carries the provider name plus any provider-specific credentials needed to query its model catalog.- Python SDK
- REST API
Save User Config
Persist user-scoped configuration on the server (for example UI preferences).- Python SDK
- REST API
ReturnsA string status or message from the server.
Load User Config
Load user-scoped configuration from the server.- Python SDK
- REST API
Named LLM Connections
Beyond the single LLM configured throughLLM_PROVIDER and friends, you can
save any number of named connections — one per provider, model, or
credential you want to switch between. An llm concept can then name the
connection it runs against instead of carrying credentials of its own.
Connections live in your user config under llm_connections, but edit them
through the endpoints below rather than through save-config: save-config
merges only at the top level, so writing the whole map from a copy you read
earlier silently drops any connection saved in between.
These endpoints are REST-only — the Python SDK does not wrap them.
Save a Connection
Creates a connection, or updates it in place if the name already exists.Request body
settings accepts llm_provider, llm_version, llm_api_key,
azure_llm_endpoint, azure_llm_api_version, databricks_workspace_url,
databricks_client_id, databricks_client_secret and
snowflake_account_url. Any other key is dropped.
Response
The non-secret summary of what was saved:The two credential fields —
llm_api_key and databricks_client_secret — are
never returned to a client, so they follow a different update rule from the
rest: omitting one leaves the stored value unchanged, while sending it
explicitly empty clears it. Every other field is taken exactly as sent, so
switching a connection’s provider drops the previous provider’s endpoints. This
is what lets an edit form save a connection it can’t display the key for.Delete a Connection
Removes one connection, leaving the others alone.data.removed is false when there was no such connection — deleting a name
that does not exist is not an error.
Get Effective LLM Defaults
Returns the provider, model and temperature an LLM call actually runs with when nothing overrides them — your settings merged over the platform defaults — plus the per-provider default models and the token and row budgetsllm concepts use.
Nothing secret is returned: no keys, no endpoints.
Use it to show real effective values in a UI instead of an opaque “default”.
Response
connections is the same non-secret summary of your saved connections, included
here so a model picker can be rendered without a second request.

