Chat Completions

The /v1/chat/completions endpoint serves unified OpenAI- and Anthropic-compatible chat completions across provider clouds at exact provider list prices with zero percent (0%) markup. The request and response shape adheres to the OpenAI Chat Completions API specification with enterprise reliability, safety, and telemetry controls.

Key capabilities:

• Unified Wire: Routes across OpenAI, Anthropic, Google Vertex AI, Azure Foundry, and Alibaba US using a single virtual key (sk-nrouter-*).
• 4-Phase Preflight: Validates key ACLs, checks rate limits (RPM/TPM), executes concurrent content moderation and prompt injection defense, and reserves credit before upstream dispatch.
• Direct & Automatic Selection: Target a specific model (Direct) or use nrouter/auto, with custom per-request failover chains (nrouter_fallbacks).
• Edge Telemetry: Returns real-time latency, upstream routing, and exact spend via x-nr-* headers (x-nr-request-cost, x-nr-latency-ms, x-nr-request-id).

POST https://api.nrouter.ai/v1/chat/completions

# Point Claude Code to nRouter gateway
export ANTHROPIC_BASE_URL="https://api.nrouter.ai"
export ANTHROPIC_API_KEY="$NROUTER_API_KEY"

# Launch with gpt-5.4-mini
claude --model gpt-5.4-mini

Examples

import { nRouter } from "@nrouter_ai/sdk";
const client = new nRouter({ apiKey: process.env.NROUTER_API_KEY });
const res = await client.chat.completions.create({
  model: "gpt-5.4-mini",
  messages: [{ role: "user", content: "Hello!" }],
});
console.log(res.choices[0].message.content, client.lastResponse?.cost);

Streaming & Routing

Pass "stream": true to stream chunks formatted as Server-Sent Events (data: {...}). Latency (x-nr-latency-ms) measures edge Time-To-First-Byte (TTFB); spend settles atomically at end-of-stream.

Two ways to ask for a model in the "model" field:

Smart Router aliases were withdrawn on 2026-10-15; a request naming one returns the unknown-model error.

Request Options

Optional body fields configure single-call routing, guardrails, and caching (stripped before upstream forwarding):

curl -X POST https://api.nrouter.ai/v1/chat/completions \
  -H "Authorization: Bearer $NROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.4-mini","messages":[{"role":"user","content":"Hello!"}],"nrouter_fallbacks":["claude-sonnet-4-5"],"nrouter_guardrails":["pii-strict"],"nrouter_cache":false}'

See Per-Request Options and Guardrails.

Headers Reference

Inbound: Authorization: Bearer sk-nrouter-... and Content-Type: application/json required; optional: x-nr-tags, x-nr-compress, x-nr-mcp-server.

Prompt Compression & Guardrails

Prompt compression is enabled by default for qualified models (opt out with x-nr-compress: off or "nrouter_compress": false). Preflight executes (1) key validation, (2) rate limits (RPM/TPM), (3) content inspection, and (4) credit reservation. Blocked requests return HTTP 400 (x-nr-guardrails: blocked) with $0 spent and $0 held. See Guardrails.

curl -X POST "https://example.com/v1/chat/completions" \  -H "Content-Type: application/json" \  -d '{    "model": "gpt-5.4-mini",    "messages": [      {        "role": "system",        "content": "You are a helpful AI assistant powered by nRouter."      },      {        "role": "user",        "content": "What models does nRouter support?"      }    ],    "temperature": 0.7,    "max_tokens": 1024  }'

{  "id": "chatcmpl-abc123",  "object": "chat.completion",  "model": "gpt-4o",  "choices": [    {      "index": 0,      "message": {        "role": "assistant",        "content": "Hello!"      },      "finish_reason": "stop"    }  ],  "usage": {    "prompt_tokens": 9,    "completion_tokens": 2,    "total_tokens": 11  }}