SDK Configuration

Timeouts

By default, the Cencori SDKs have a 60-second read timeout. You should increase this for long-running chain-of-thought models (like o1 or claude-opus).

TypeScript

const cencori = new Cencori({
  apiKey: '...',
  timeout: 300000 // 5 minutes (in ms)
});

Python

client = Cencori(
    timeout=300.0 # 5 minutes (in seconds)
)

Go

client := cencori.NewClient(
    option.WithHTTPClient(&http.Client{
        Timeout: 5 * time.Minute,
    }),
)

Retries

The SDKs automatically retry failed requests (status 408, 409, 429, 5xx) with exponential backoff.

• Default Retries: 2
• Max Delay: 60 seconds

To customize:

const cencori = new Cencori({
  maxRetries: 5 // Retry up to 5 times
});

Edge Compatibility (Vercel / Cloudflare)

The Cencori TypeScript SDK (npm install cencori) is fully compatible with Edge Runtimes.

• Zero dependencies on Node.js-only APIs (like fs or net).
• Uses the standard fetch API.
• Works in Cloudflare Workers, Vercel Edge Functions, and Deno.

// pages/api/edge.ts
export const config = {
  runtime: 'edge',
};
 
export default async function handler(req) {
  // Works natively in Edge
  const response = await cencori.ai.chat(...) 
  return new Response(...)
}

Telemetry & Privacy

The Cencori SDK collects minimal, anonymous metadata to help us improve performance and debug issues.

We Collect:

• SDK Version (e.g., 1.2.0)
• Language & Runtime Version (e.g., Node v18.16.0, Python 3.11)
• Platform OS (e.g., darwin, linux)
• Latency (Client-side measured duration)

We Do NOT Collect:

• Environment variables (other than CENCORI_API_KEY)
• File system paths
• IP Addresses (these are visible to the gateway, but not collected by the SDK logic itself)

To opt-out of telemetry headers:

export CENCORI_TELEMETRY=0

Web Telemetry (App Traffic)

Separately from anonymous SDK telemetry, you can report web traffic from your app to the Cencori dashboard. This populates the Logs → Web Gateway tab with your app's real domain.

// In your middleware or API route handler:
await cencori.telemetry.reportWebRequest({
  host: req.headers.get('host') || 'unknown',
  method: req.method,
  path: new URL(req.url).pathname,
  statusCode: response.status,
  latencyMs: Date.now() - startTime,
});

This is fire-and-forget — it never throws and never blocks your application.