Subprocessor disclosure
Meridian routes inference, billing, and observability through a curated set of upstream vendors. This page enumerates every subprocessor that may touch customer prompts, completions, or account metadata, plus the contractual posture under which they operate.
1.Inference providers
Prompts you send to /v1/chat/completions are forwarded to one or more model hosts depending on the model alias you request. Each upstream is bound by a zero-retention DPA where available.
- Microsoft Azure OpenAI (Sweden Central, East US 2)
- Anthropic via Azure-bridged endpoints
- Cohere, DeepSeek, and Llama hosts behind the Azure gateway
2.Billing and identity
Stripe processes card data and stores customer billing records. Clerk handles authentication, session tokens, and organization membership. Neither vendor sees prompt or completion bodies. API keys are scoped per-organization and rotate without re-billing.
3.Logging and metrics
Request metadata (timestamps, token counts, latency, status codes) is stored in Upstash Redis for usage analytics. Prompt and completion bodies are never written to durable logs. A typical metric record looks like:
{
"org_id": "org_2abc",
"model": "azure/model-router",
"prompt_tokens": 412,
"completion_tokens": 988,
"latency_ms": 1842,
"status": 200,
"ts": "2026-06-27T18:04:11Z"
}