OpenRouter Integration
Trace every model routed through OpenRouter. wrapOpenRouter() captures which model was actually served, cost, latency, and token usage — across 100+ models.
OpenRouter is an API gateway that routes requests to 100+ LLMs from OpenAI, Anthropic, Google, Meta, Mistral, and others through a single OpenAI-compatible endpoint.
Getting Started
Install the SDK
Install @zespan/sdk — OpenRouter uses an OpenAI-compatible API.
npm install @zespan/sdk openaiWrap the OpenRouter client
Configure an OpenAI client pointing at OpenRouter's base URL and pass it to wrapOpenRouter().
import OpenAI from 'openai';
import { Zespan, wrapOpenRouter } from '@zespan/sdk';
const lt = new Zespan({ apiKey: process.env.ZESPAN_API_KEY });
const openrouter = wrapOpenRouter(
new OpenAI({
baseURL: 'https://openrouter.ai/api/v1',
apiKey: process.env.OPENROUTER_API_KEY,
}),
lt
);
const res = await openrouter.chat.completions.create({
model: 'anthropic/claude-3-5-sonnet',
messages: [{ role: 'user', content: 'Hello' }],
});View traces in Zespan
Open Trace Explorer. Each trace shows the actual model served, token usage, and USD cost with OpenRouter's per-model pricing.
What's captured automatically
- Actual model served: captures the routed model, not just the requested one
- 100+ models: GPT-4o, Claude 3.5, Llama 3, Gemini, Mistral, and more
- Per-model cost: OpenRouter's pricing applied per call
- Fallback tracking: if OpenRouter reroutes to a fallback model, it's recorded
- Standard OpenAI-compatible response tracing
FAQ
Does Zespan capture which model OpenRouter actually served?
Yes. wrapOpenRouter() reads the model field from OpenRouter's response, so the trace always shows the actual served model, not the requested one.
Start for free — 10K traces/month, no card needed
OpenRouter integration works on all plans including the free tier.