Opsmeter Integration Examples

Provider changes, Opsmeter payload stays the same.

Working examples for sending telemetry to POST /v1/ingest/llm-request in:

• .NET (examples/dotnet)
• Node.js (examples/node)
• Python (examples/python)

This repo is optimized for teams implementing LLM cost tracking, OpenAI usage monitoring, Anthropic usage telemetry, and AI inference cost control with a consistent request schema.

Opsmeter product site: https://opsmeter.io
Opsmeter API base: https://api.opsmeter.io Provider + catalog model names: https://opsmeter.io/docs/catalog Official SDK package identities for opsmeter.io:

• Node (npm): @opsmeter.io/node
• Python (PyPI): opsmeter-io-sdk

Current provider support in examples: OpenAI and Anthropic only.

This repository targets LLM telemetry quickstart, OpenAI cost tracking examples, Anthropic integration examples, and AI cost observability setup keywords.

What this repo solves

• No-proxy telemetry: keep your provider call path untouched, send attribution metadata after each LLM call.
• Retry-safe ingestion: reuse externalRequestId on retries to prevent duplicate request rows.
• Cost attribution dimensions: keep endpointTag and promptVersion consistent for feature-level and version-level analysis.

Documentation paths

• Direct ingest (current production path): Quickstart + payload contract + language examples under examples/.
• SDK auto-instrumentation (preview path): moved to dedicated SDK repositories.
• Direct ingest docs stay valid; SDK docs are additive and do not replace existing integration docs.

Official package identity (opsmeter.io)

• Official domain and product identity: https://opsmeter.io
• Official Node package name: @opsmeter.io/node
• Official Python package name: opsmeter-io-sdk
• Model catalog for both SDKs: https://opsmeter.io/docs/catalog

Table of contents

• Quickstart (60s)
• Documentation paths
• Official package identity (opsmeter.io)
• Payload contract (shared)
• Allowed values
• Recommended combinations
• Architecture
• Quick visual
• Examples
• Example modes
• Launch-ready wedges
• SDK preview
• n8n templates
• Common mistakes
• CI / Quality gates
• Product linking text (for docs/pricing/landing)
• Release
• SEO and discoverability notes

Quickstart (60s)

1. Clone and set your API key.

git clone https://github.com/opsmeter-io/opsmeter.io-integration-examples.git
cd opsmeter-integration-examples
export OPSMETER_API_KEY="<YOUR_WORKSPACE_PRIMARY_API_KEY>"
export OPSMETER_API_BASE_URL="https://api.opsmeter.io"

2. Run one stack (Node shown below):

# Provider/model names: https://opsmeter.io/docs/catalog
node examples/node/index.mjs --provider openai --model gpt-4o-mini --retry

3. Expected output:

Business call completed.
Telemetry dispatched (non-blocking).
Ingest response: 200 ok=true planTier=Free warnings=0
Retry with same externalRequestId sent.

4. Verify in product:

• Dashboard request count increases.
• endpointTag and promptVersion appear in Top Endpoints / Prompt Versions.

--retry uses the same externalRequestId to demonstrate retry-safe behavior.

Payload contract (shared)

Canonical ingest endpoint: https://api.opsmeter.io/v1/ingest/llm-request

All examples send this same shape:

{
  "externalRequestId": "ext_123abc",
  "provider": "openai",
  "model": "gpt-4o-mini",
  "promptVersion": "summary_v3",
  "endpointTag": "checkout.ai_summary",
  "inputTokens": 120,
  "outputTokens": 45,
  "totalTokens": 165,
  "latencyMs": 820,
  "status": "success",
  "errorCode": null,
  "userId": null,
  "dataMode": "real",
  "environment": "prod"
}

provider and model values should be selected from the catalog: https://opsmeter.io/docs/catalog

Allowed values

Field	Allowed	Notes
provider	openai, anthropic	Current support in this repo/examples
status	success, error	Required by API validation
dataMode	real, test, demo	Default recommendation: real
environment	prod, staging, dev	Use real deployment environment

Recommended combinations

Use case	dataMode	environment
Production traffic	real	prod
QA/Test traffic	test	staging or dev
Seed/demo flows	demo	dev

If you do not label these fields correctly, dashboard analytics can mix operational and non-production signals.

Architecture

flowchart LR
  A["LLM call"] --> B["Map usage + latency"]
  B --> C["Build Opsmeter payload"]
  C --> D["POST /v1/ingest/llm-request"]
  D --> E["Dashboard / Budgets / Alerts"]

Loading

Quick visual

Examples

• Node examples (without SDK + with SDK)
• Python examples (without SDK + with SDK)
• Dotnet examples (without SDK + with SDK)

Launch-ready wedges

These are lighter assets optimized for GitHub discovery, founder distribution, and quick proof-of-value sharing.

Important framing:

• The payload contract stays provider-agnostic.

• These folders are provider-specific entry points for faster evaluation.

• The generic direct-ingest examples under examples/node, examples/python, and examples/dotnet remain the canonical shared pattern.

• OpenAI cost tracker example (Node)

  • Real OpenAI call + usage extraction + direct-ingest telemetry
  • Best first asset for JavaScript-heavy product teams evaluating AI cost tracking with minimal setup

• Anthropic cost tracker example (Node)

  • Real Anthropic Messages API call + usage extraction + direct-ingest telemetry
  • Keeps the same Opsmeter payload shape while changing only the provider call

• OpenAI cost tracker example (Python)

  • Same wedge for Python-heavy AI backends and internal tools
  • Useful when the evaluation owner is closer to ML or backend workflows

• Anthropic cost tracker example (Python)

  • Same provider-specific entry point for Python teams using Claude
  • Makes the generic telemetry pattern obvious across providers

Example modes

• Without SDK (direct ingest): existing stable production path in this repo.
• Includes explicit send scenarios for both OpenAI and Anthropic.
• With SDK (preview): usage examples in this repo, SDK packages in dedicated repos:
  • Includes OpenAI + Anthropic capture/send scenarios in language samples.
  • Node SDK repo: github.com/opsmeter-io/opsmeter.io-node-sdk
  • Node npm package (published): npmjs.com/package/@opsmeter.io/node
  • Python SDK repo: github.com/opsmeter-io/opsmeter.io-python-sdk
  • Python package (published): pypi.org/project/opsmeter-io-sdk
  • .NET SDK repo: coming soon

SDK preview

Preview SDK contracts and reference implementations are maintained in dedicated repositories:

• Node SDK (repo): github.com/opsmeter-io/opsmeter.io-node-sdk
• Node SDK (npm): npmjs.com/package/@opsmeter.io/node
• Python SDK (repo): github.com/opsmeter-io/opsmeter.io-python-sdk
• Python SDK (PyPI): pypi.org/project/opsmeter-io-sdk
• .NET SDK (repo): coming soon
• .NET package: coming soon

n8n templates

These templates are for Opsmeter n8n integration with workspace status branching, budget warning automation, and telemetry paused handling.

Path: ./n8n

• workspace-status-check.json: polls GET /v1/diagnostics/workspace-status and branches by plan/budget booleans.
• budget-warning-to-slack.json: scheduled budget status check with Slack notification path.
• openai-to-opsmeter-ingest.json: provider call + usage mapping + ingest + 402 plan-limit branch.
• Import and setup guide: n8n/README.md

Common mistakes

Common mistakes

• Provider/model typo (example: wrong provider string), causing unknown model attribution.
• Generating a new externalRequestId for retries (breaks idempotent behavior).
• Blocking the request path with long telemetry timeouts.
• Treating telemetry failure as business failure (it should be swallowed/logged).

CI / Quality gates

• Node lint + tests
• Python lint + tests
• Dotnet build + tests
• Smoke script runs all three examples in dry-run mode

See .github/workflows/ci.yml and scripts/smoke.sh.

Product linking text (for docs/pricing/landing)

Use this exact label when linking from the main product:

Integration examples (60-second quickstart)

Target URL:

https://github.com/opsmeter-io/opsmeter.io-integration-examples

Release

Current bootstrap release target: v0.1.0 (see CHANGELOG).

SEO and discoverability notes

Primary terms covered in this repository:

• Opsmeter integration examples
• LLM cost tracking integration
• OpenAI usage monitoring
• Anthropic telemetry integration
• AI inference cost control