Model-Based Simulation for Validation

Explore inference cost, latency, and evidence needs.

This separate Playground supports model-based cost and latency scenarios and, when credentials are configured, one direct provider call for a single request. The separate /api/inference endpoint returns simulated decision receipts with illustrative route metadata; it does not call a provider or route production workloads. For a configured Playground call, cost is estimated from returned token usage and disclosed catalog pricing, and latency is measured for that request only. Environmental impact data remains unavailable.

Without configured provider credentials, the Playground returns a deterministic worked example: model output and token counts are simulated, cost comparison is unavailable, and latency is illustrative or unavailable. With credentials, it can return one direct provider response marked LIVE only for that provider call. Region targets are illustrative. No result provides live workload routing, production telemetry, measured environmental impact, or a verified combined impact receipt.

One request, one receipt
POST the same shape to /v1/chat/completions. This demo is non-streaming and never persists prompts or receipts.
server-side keys

One user message is sent. Combined message content is capped at 30,000 characters.

Provider selection for one request
Without provider credentials, the playground returns simulated output. A configured provider receives one direct request.
Effective safety settings
Loaded from the server. These limits are process-local and in-memory for this no-persistence MVP.
Rate limit
5/IP/60s
CARBONLAYER_RECEIPT_RATE_LIMIT_MAX
Global spend cap
$1.00 lifetime
CARBONLAYER_RECEIPT_GLOBAL_SPEND_CAP_USD
Rate window
60s rolling per-IP window
CARBONLAYER_RECEIPT_RATE_LIMIT_WINDOW_SECONDS