Skip to main content
Connecting your app is a base URL swap. You keep your provider, your SDK, and your application logic. Catalyst sits in the middle, captures everything, and aggregates metrics across all your LLM providers in one place.

​
The quick change

Point your SDK at Catalyst. Your project API key authenticates the request, and your provider key goes in x-inference-provider-api-key so the gateway can forward it. The example below uses OpenAI. For other providers see the Integrations guide. 
import OpenAI from "openai";

const client = new OpenAI({
baseURL: "https://api.inference.net/v1",
apiKey: process.env.INFERENCE_API_KEY,
defaultHeaders: {
"x-inference-provider-api-key": process.env.OPENAI_API_KEY,
"x-inference-provider": "openai",
},
});

const response = await client.chat.completions.create({
model: "gpt-4.1",
messages: [{ role: "user", content: "Hello, world!" }],
});
That’s it. Every request now flows through Catalyst and gets captured.

​
Prefer AI-assisted setup

Prefer to avoid manual edits? Use Install with AI. It runs inf instrument, scans your codebase, updates your LLM clients to use the gateway, and adds task IDs automatically.

​
What gets captured

Once traffic is flowing, every request records:
  • Full request and response payloads
  • Cost (per call and aggregate)
  • Latency (end-to-end and time to first token)
  • Token counts (input and output)
  • Cache hit rates
  • Error rates and status codes
  • Model and provider information
  • Function/tool call details
  • Whether the request includes images
  • Any task tags you set
Catalyst works with any OpenAI-compatible provider including Anthropic, Groq, Cerebras, and OpenRouter. See the full integration guides for provider-specific setup.

​
What happens next

Your data shows up immediately in two places:

Metrics Explorer

Dashboards for cost, latency, errors, and usage aggregated across all your providers.

Inference Viewer

Browse and filter individual LLM requests and responses.
Already connected? Set up tasks to group calls by objective. This is what powers per-feature metrics, evals, and training.