Know what caused the spend.
Break usage down by app, model, feature, environment, user, trace, and session.
AI telemetry SDK / 01
Every token. Accounted for.
Instrument your AI applications with exact token, cost, latency, and feature attribution—without proxying traffic or surrendering provider credentials.
0prompts stored
<100msingestion target
6typed adapters
SDK setup / 02
From model response to cost insight in minutes.
One server-side SDK captures provider-reported usage and the business context that a machine-level command cannot reliably infer.
Create your account →- 01
Create an app key
Create a Tokly workspace and app, then save the server-only write key shown once.
- 02
Install the SDK
Add the lightweight TypeScript package to the application that calls your model provider.
npm install tokly-sdk - 03
Capture real usage
Send provider usage through a typed adapter and attach the feature and environment that caused it.
tokly.capture(fromOpenAIResponse({ model, usage })); - 04
Operate with context
Inspect exact tokens, attributed cost, latency, pricing coverage, and budget thresholds in Tokly.
AdaptersVercel AIOpenAIAnthropicGeminiOpenRouterOllama
Ready-to-use code
Choose your provider example.
Copy the complete server-side example, add your app key as an environment variable, and capture the usage returned by your provider.
Pass the model and usage from an OpenAI Responses result.
import { createTokly } from "tokly-sdk";
import { fromOpenAIResponse } from "tokly-sdk/adapters";
const tokly = createTokly({
apiKey: process.env.TOKLY_API_KEY!,
});
tokly.capture(fromOpenAIResponse({
model: response.model,
usage: response.usage,
dimensions: {
environment: "production",
feature: "assistant",
},
}));
await tokly.flush();Run Tokly only on your server. Keep TOKLY_API_KEY in an environment variable and flush before a serverless request ends.
Price history, not guesswork.
Effective-dated model rates preserve the cost calculation that applied when each call ran.
Budgets that speak up early.
Threshold alerts reach your team before a monthly AI bill becomes an incident.