Straight answers. Clear expectations.

Model Meter is a developer tool that gives engineering teams visibility into their AI spending. It tracks every LLM API request across multiple providers, attributes costs to specific projects or features, and displays spend trends on a live dashboard — all with a two-line SDK integration.

No. Model Meter is a cost tracking and visibility tool, not a proxy or gateway. It does not sit in the critical path of your AI calls. The SDK instruments existing API calls via OpenTelemetry and sends telemetry to Model Meter asynchronously. Your AI requests go directly to the provider.

No. Model Meter does not collect prompts or completions by default. The SDK only captures: provider and model name, token usage (input, output, cached), latency, success or error status, operation label (optional, set by you), and cost estimate. No prompt text, response text, or personally identifying data is transmitted.

SDK integration in an existing Node.js app typically takes 10–15 minutes: install the package, call init(), and set up the OTel span processor. After that, every generateText() and streamText() call is tracked automatically.

The Model Meter SDK is on GitHub — install it and call init() once at startup. The hosted product will be a paid subscription; pricing hasn't been finalized. Join the waitlist to lock in the best launch pricing before public rates go live.

Verified providers: OpenAI (GPT-4o, GPT-4o-mini, GPT-4 Turbo, GPT-3.5), Anthropic (Claude 3.5 Sonnet, Claude 3.5 Haiku, Claude 3 Opus), Google Gemini (1.5 Pro, 1.5 Flash, 2.0 Flash), Mistral, Groq, and Cohere. The SDK captures usage from any provider supported by the Vercel AI SDK via OpenTelemetry gen_ai.* spans.

They are estimates, not exact billing amounts. Provider pricing changes, discounts, credits, batch rates, and surcharges are not always captured by request-level estimates. Model Meter marks estimates clearly as 'estimated' vs. 'provider-reported' vs. 'reconciled'. The billing reconciliation feature lets you import provider invoices and compare them to estimates.

Model Meter uses effective-dated pricing records. Each cost record stores the pricing version that was active at the time of the request, so historical cost data is not silently rewritten when prices change. New pricing records are added with an effective start date.

Install the SDK and call init() once at startup — telemetry flows to your Model Meter workspace asynchronously, outside the critical path of your AI calls. A managed cloud product is coming; join the waitlist for the best launch pricing.

Yes. Model Meter integrates with the Vercel AI SDK via OpenTelemetry. After calling init(), set up the OTel provider with createModelMeterExporter and call registerTelemetry(new OpenTelemetry()). All subsequent generateText(), streamText(), generateObject(), and embed() calls are tracked automatically.

Both tools track LLM usage and cost. Key differences: Model Meter instruments your app via the SDK and does not act as a proxy. Helicone routes your calls through a proxy service. Model Meter does not store prompts or completions by default; Helicone stores request/response data by default. Choose Model Meter when you cannot route traffic through a third-party proxy, or when cost attribution by project is the primary need.

LangSmith is primarily a prompt engineering and evaluation platform (tracing, annotation, datasets, evals). Model Meter is primarily a cost visibility and spend attribution tool. They address different problems and can be used together: LangSmith for prompt debugging during development, Model Meter for production cost tracking and finance reporting.

Read CONTRIBUTING.md in the GitHub repo, then browse 'good first issue' labels. Good starting points: adding or verifying provider pricing fixtures, writing tests, or improving documentation. Open an issue before large changes so scope and design can be discussed.