Pay-i Databricks (Proxy) icon

Pay-i Databricks (Proxy)

Databricks Model Serving chat model routed through Pay-i proxy for cost tracking and budget enforcement

Pay-i Databricks (Proxy)

AI-generated

Summary

Supplies a LangChain ChatOpenAI language model that sends chat requests to a Databricks Model Serving endpoint through the Pay-i proxy, so usage is tracked for cost attribution and limit enforcement. It is a sub-node that connects to an AI Agent or Chain through the Model input.

Inputs

  • endpointName (required) — Databricks serving endpoint, chosen from a searchable list (fetched from the workspace) or entered by name. It is used as the model name in requests.
  • deployedModel — The underlying model behind the endpoint, chosen from a list of Pay-i resources or entered by name. It is sent as the price-as resource. Required for custom endpoints whose names don't start with "databricks-"; otherwise the node throws an error.
  • cloudProvider (required) — AWS, Azure or Google Cloud. Selects the Pay-i pricing category (system.databricks.).
  • options — Sampling temperature, max tokens (-1 for no limit), Top P, frequency penalty and presence penalty. These are passed to the ChatOpenAI model.
  • userId — User identifier for cost attribution.
  • useCaseName — Pay-i use case name. Defaults to the workflow name with spaces replaced by hyphens.
  • useCaseId — Use case instance ID. Defaults to the n8n execution ID, so each run gets its own Pay-i use case.
  • useCaseStep — Step label. If left at the default "Pay-i Databricks (Proxy)" or empty, the node's canvas name is used instead.
  • advancedTracking — Use case version, use case properties (JSON), limit IDs (comma-separated), Flatten Gemini Content (default on) and Debug Logging (default off, very verbose). Change these only on Pay-i Support's guidance.

Output shape

A LangChain ChatOpenAI model object on the AiLanguageModel connection (output "Model"). It makes no direct data output; downstream AI nodes call it.

Requires two credentials: payiApi (base URL and API key) and payiDatabricksApi (workspace URL and access token). Requests go to {Pay-i base URL}/api/v1/proxy/openai/v1 with xProxy-* headers: API key, Databricks provider base URI, price-as category and resource, and the tracking fields. Header values are sanitized to ASCII. When Flatten Gemini Content is on, array content blocks in responses are collapsed into a plain text string. Debug Logging writes request details to the n8n server log.

Examples

Example 1: Use a Databricks pre-provisioned foundation model in an AI Agent with cost tracking

Serving Endpoint Name: databricks-gpt-5-4 (from list); Cloud Provider: AWS; leave tracking fields at defaults; connect Model output to the AI Agent.

Example 2: Use a custom serving endpoint and attribute cost to a user and step

Endpoint: my-custom-endpoint; Deployed Model: the matching Pay-i resource (required); Cloud Provider: Azure; User ID: an expression for the user; Use Case Step: "Step 1 - Outline".

Links

Discussion