The Modelmetry SDK provides a JS/TS interface to interact with the Modelmetry API, allowing developers to easily integrate Modelmetry's guardrails and observability features in their codebases.
To get started with the Modelmetry SDK, you first need to install it. You can do this using pip:
npm install @modelmetry/sdk
We recommend you look at the ./examples
directory for actual examples.
Here's a quick example to show you how to instantiate the SDK client and perform a check using the Modelmetry API.
Initialize the SDK
import { ModelmetryClient } from "../src/sdk"
import 'dotenv/config'
const modelmetry = new ModelmetryClient({
tenantId: process.env.TENANT_ID, // required
apikey: process.env.API_KEY, // required
baseUrl: HOST, // optional
})
Perform a guardrail check
export const basicGuardrailExample = async () => {
const response = await openai.chat.completions.create({ max_tokens: 500, model, messages });
const responseText = jokeResponse.choices[0].message.content;
const result = await modelmetry.guardrails().check("grd_jaohzgcbd5hbt1grwmvp", {
Input: {
Text: {
Text: responseText,
},
},
})
if (result.failed) {
// Handle a failed check
console.error("Failed check", result)
}
if (result.errored) {
// Handle an errored check (an error means an unexpected error occurred, not that the check failed).
// By default, an error results in a "passed" check with this errored property set to true.
console.warn("Errored check", result)
}
// The check passed, carry on
console.info("Passed check", result)
}
await basicGuardrailExample();
Use observability
fastify.get("/joke", async function handler(request, reply) {
// create a new trace
const trace = modelmetry.observability().newTrace("get-joke");
try {
// child span for the first task
const spanTask1 = trace.span("task-1", "other", {});
const stuff = await runTheFirstTask();
spanTask1.end();
// child span for the second task, this time we pass the trace to the task
// so that it can create a child span itself
const secondStuff = await runTheSecondTask(stuff, trace);
reply.send({ secondStuff });
} catch (err) {
reply
.status(500)
.send({ error: "Failed to fetch joke", message: err.message });
} finally {
trace.end();
}
});
function runTheSecondTask(stuff, trace) {
// this time, use a callback to create a child span's scope,
// and it will end automatically when the callback returns
return trace.startSpan("task-2", "other", async (span) => {
span.setAttribute("user", "user-123");
span.setAttribute("stuff", stuff);
// do something
return "second stuff";
});
}
To use the Modelmetry SDK, you must authenticate using your tenant ID and API key. You can find these in your Modelmetry settings.
When creating the ModelmetryClient
instance, pass your tenant_id
and api_key
as shown in the Quick Start example above. These credentials will be used for all API calls made through the SDK client.
Modelmetry provides advanced guardrails and monitoring for applications utilizing Large Language Models (LLMs).
Modelmetry offers tools to prevent security threats, detect sensitive topics, filter offensive language, identify personally identifiable information (PII), and ensure the relevance and appropriateness of LLM outputs.
Modelmetry’s platform integrates with leading AI providers, allowing developers to customize evaluators for enhanced safety, quality, and compliance in their AI-driven solutions.