🛠️Integration Options for AI Metering

The latest list of integrations will always be available from inside the Revenium application when logged in.

Overview

Revenium provides middleware libraries and integration options to simplify metering across a wide range of AI frameworks, languages, and model providers. These integrations allow developers to automatically track token usage, costs, latencies, and metadata without rewriting core application logic.

Each middleware integration captures relevant details from the AI provider’s response and reports them to Revenium’s usage API. This enables immediate access to full observability and billing insights via the AI Analytics and Alerts dashboards.

Below is a summary of currently supported middleware integrations across languages, SDKs, and providers.

---

✅ Supported Middleware Integrations

Language	AI or Proxy Provider	Chat Completions	Embeddings
python	OpenAI	✅	✅
python	Azure OpenAI	✅	✅
python	liteLLM	✅	✅
python	ollama	✅	🚧 In-Progress
python	Anthropic	✅	Anthropic has no embedding models
python	Anthropic via aws bedrock	✅	Anthropic has no embedding models
python	Google Vertex SDK (Enterprise)	✅	✅
python	Google ai SDK	✅	✅
python + Langchain extension	openAI	✅	✅
nodeJS	OpenAI	✅	✅
nodeJS	Azure OpenAI	✅	✅
nodeJS	liteLLM Proxy	✅	✅
n8n – Chat Agent	OpenAI	✅	✅
python – griptape driver	OpenAI	✅	✅

---

ℹ️ If you use a different SDK or framework not listed here, Revenium also supports direct API calls to our metering endpoint. You can forward usage data from any custom application or proxy with minimal integration effort.

📩 Contact [email protected] to request a custom SDK, integration assistance, or to contribute a plugin to the open middleware ecosystem.

Usage Metadata Reference

The usage_metadata object is used to pass additional fields to Revenium for billing, cost attribution, analytics, and alerting. These fields are optional, but the more you provide, the more granular and useful your reporting will be.

⚠️ Naming conventions: Field names are shown here in camelCase, which is recommended for JavaScript, TypeScript, and JSON-based integrations. For Python or other ecosystems, use idiomatic casing (e.g. snake_case in Python).

---

Top-Level Fields

Field	Description	Use Case
traceId	Unique ID for a session or conversation	Group multi-turn or batch requests for unified cost tracking
taskType	High-level classification of the AI task (e.g., summary, chat)	Attribute cost to types of work (summarization, retrieval, etc.)
organizationId	Customer or department ID from your systems	Track usage by external customer or internal business unit
subscriptionId	Identifier for billing plan or contract	Used for usage-based billing and quota enforcement
productId	Your internal product name or feature label	Group usage and cost by monetized product
agent	LLM identifier or orchestration wrapper name (e.g., gpt-4-summary)	Analyze agent-level performance or cost
responseQualityScore	Numeric score (0.0–1.0) indicating quality of the AI response	Measure perceived or model-generated quality
subscriber	See below — structured object defining the user or process that made call	Enables user-level attribution and grouping

---

Subscriber Object

The subscriber field can be a nested object to help organize and identify individual users or systems making the request.

Example

"subscriber": {
  "id": "[email protected]",
  "email": "[email protected]",
  "credential": {
    "name": "EHR Key – Nurse Station A",
    "value": "pk-ehra57"
  }
}

Fields

Field	Description
id	Unique internal identifier for the subscriber
email	Email address of the user (optional)
credential.name	Friendly label for the API key or access token
credential.value	Actual token used to authenticate the request

📘 Tip: For anonymous usage, you can set only the credential value. When the user is later identified, you can backfill the id or email to unify history.

---

🧠 Best Practices

• Always send organizationId and subscriber.id to unlock the full value of per-customer and per-user cost tracking.

• Use taskType, productId, and agent to group usage by product and capability for internal ROI reporting.

• Set traceId to group multi-part requests (e.g., retrieval + summarization).

• Set responseQualityScore if you have human or automatic scoring to surface low-quality responses.

---

🛠️ Example Usage Metadata Object

{
  "traceId": "session-38274",
  "taskType": "summarize-discharge-notes",
  "organizationId": "brighthealth-clinops",
  "subscriptionId": "clinical-tier-2",
  "productId": "ehr-summary",
  "agent": "gpt-4-discharge",
  "responseQualityScore": 0.94,
  "totalCost": 0.0215,
  "subscriber": {
    "id": "nurse.a57",
    "email": "[email protected]",
    "credential": {
      "name": "EHR Key – Nurse Station A",
      "value": "pk-ehra57"
    }
  }
}

---

Direct API Integration

Using one of our supported middleware options is far lower effort and automates the submission of many of the required fields for AI metering. However, if you need complete customization and control over how you integrate with Revenium, the direct API integration option exists as well.

Our API integration documentation for AI metering is available here. An example of the API structure to send metering data is shown below.

Many of the fields shown below are optional to demonstrate the possibilities for metadata enrichment. View the complete API documentation using the link above to verify required vs. optional fields.

curl -vX POST "https://api.dev.hcapp.io/meter/v2/ai/completions" \
     --header 'accept: application/json' \
     --header 'content-type: application/json' \
     --header 'x-api-key: hak_your_key' \
     --data '

{
  "stopReason": "END",
  "costType": "AI",
  "isStreamed": true,
  "taskType": "discharge-summary",
  "agent": "clinical-summary-agent",
  "operationType": "CHAT",
  "inputTokenCount": 200,
  "outputTokenCount": 500,
  "reasoningTokenCount": 0,
  "cacheCreationTokenCount": 0,
  "cacheReadTokenCount": 0,
  "totalTokenCount": 700,
  "organizationId": "brighthealth-clinic",
  "productId": "ehr-integration",
  "subscriber": {
	  "id": "subscriberId-nurse-a57",
	  "email": "[email protected]",
	  "credential": {
	    "name": "EHR Access Token",
	    "value": "pk-ehra57"
	    }
  },
  "model": "gpt-3.5-turbo",
  "transactionId": "4yU2vUKQQE2lv7TWB6F3U5C05at5J2D",
  "responseTime": "2025-06-16T22:35:39.592Z",
  "requestDuration": 2400,
  "provider": "OpenAI",
  "requestTime": "2025-06-16T22:35:34.592Z",
  "completionStartTime": "2025-06-16T22:35:39.592Z",
  "timeToFirstToken": 10200
}
'