nrouter 2.1.1

nRouter SDK — one API key for models across six provider clouds. Works in Flutter and plain Dart.

nRouter SDK for Dart & Flutter

One API key for models across six provider clouds. One dependency (http), so the same code runs on Flutter mobile, desktop, web, and the plain Dart VM.

dependencies:
  nrouter: ^2.1.1

Use it

import 'package:nrouter/nrouter.dart';

final client = NRouter(apiKey: myKey);

final result = await client.chatCompletions({
  'model': 'claude-sonnet-4-5',
  'messages': [{'role': 'user', 'content': 'Hello!'}],
});

print(result.body['choices']);
client.close();

Streaming

The four text-generation wires expose cold, cancellable SSE streams. The SDK sets stream: true, accepts OpenAI [DONE] and native Anthropic message_stop terminators, and turns in-band gateway errors into the same typed exceptions as buffered calls.

await for (final chunk in client.messagesStream({
  'model': 'claude-haiku-4-5-20251001',
  'max_tokens': 64,
  'messages': [{'role': 'user', 'content': 'Hello!'}],
})) {
  print(chunk.delta);
}

Available helpers are chatCompletionsStream, completionsStream, messagesStream, and responsesStream; stream(path, body) is the generic escape hatch. Cancelling the subscription stops consuming the underlying HTTP response.

Streaming metadata is captured from the opening response headers. The final request cost is normally unknown there because the headers are sent before generation completes; cost therefore remains null, never a misleading zero.

Why there is no environment fallback

The server-side nRouter SDKs read NROUTER_API_KEY. This one deliberately does not: Platform.environment requires dart:io, which does not exist in a Flutter web build, and on mobile it is empty anyway. A fallback that quietly resolves to nothing is worse than no fallback, so apiKey is required.

Do not ship a key in the app

Anything compiled into a Flutter bundle is readable by anyone who downloads it — and on web it is served in plain text. A shipped key is a published key, and an nRouter key spends real credits. Mint a short-lived key on your backend and pass it here.

What a call cost

final meta = result.meta;
print('request ${meta.requestId} | model ${meta.model}');

// Branch on the status, never on `cost` being null-ish. An unpriced model
// reports cost == null, and rendering that as $0 reports a free request —
// which no enabled model is.
print(meta.isPriced ? 'cost \$${meta.cost}' : 'cost unpriced');

NRouterResponseMeta carries all thirteen x-nr-* headers: requestId, cost, costStatus, model, inputTokens, outputTokens, totalTokens, cacheReadTokens, cacheWriteTokens, limitSource, authReason, responseCache, responseCacheAge.

Errors

Every refusal is a subclass of the sealed NRouterError, chosen from the gateway's stable code — not the HTTP status, which cannot separate the two 400s or the two 429s. Because it is sealed, a switch over it is exhaustive and the analyzer tells you when a new case appears.

try {
  await client.chatCompletions(body);
} on NRouterGuardrailBlockedError {
  // a rule denied it; changing the request is the fix
} on NRouterCreditError {
  // out of credits; topping up is the fix
} on NRouterRateLimitError catch (e) {
  // e.body?.limitSource names WHICH ceiling. null when the gateway could not
  // attribute the refusal — this SDK does not guess, because sending a
  // customer to raise the wrong limit is worse than saying nothing.
  if (e.isRetryable) await retryLater();
}

Type	Code(s)	HTTP
NRouterRequestError	invalid_request	400
NRouterGuardrailBlockedError	guardrail_blocked	400
NRouterAuthenticationError	invalid_api_key	401
NRouterCreditError	insufficient_credits	402
NRouterNotFoundError	model_not_found	404
NRouterRateLimitError	rate_limit_exceeded, tpm_limit_exceeded	429
NRouterServiceError	credit_check_failed, service_unavailable	503
NRouterOtherError	anything newer than this SDK	—
NRouterTransportError	never reached the gateway	—

isRetryable is true only for rate-limit, service and transport failures.

Configuration

NRouter(
  apiKey: myKey,
  baseUrl: 'https://api-stage.nrouter.ai/v1',
  httpClient: myClient,        // your own http.Client — proxy, retries, mocks
);

Pass your own http.Client and this SDK will not close it; the one it creates itself is released by close().

Endpoints

All 15 gateway operations have named helpers: chatCompletions, completions, embeddings, imagesGenerations, messages, countTokens, responses, models, model, createVideo, retrieveVideo, downloadVideoContent, audioSpeech, audioTranscriptions, and audioTranslations. post, get, bytes, and multipart remain available as escape hatches.

Not JSON: audioTranscriptions and audioTranslations send multipart/form-data (the gateway requires a binary file part, so the JSON helpers cannot reach them); audioSpeech() and downloadVideoContent() return raw bytes plus metadata; bytes(path, body) remains available for other non-JSON responses. The JSON helpers refuse a non-JSON response rather than handing back an empty body for a request you were billed for.

Build and test

dart pub get
dart analyze
dart test

Publishing: PUBLISHING.md.

How guardrails, budgets and routing work

They are configured in the dashboard and enforced at the gateway, not in this package. The useful guarantee is not that they are always on — it is that whatever you have enabled cannot be bypassed by a client, this one included, and behaves identically from every nRouter SDK and from raw curl.

• Guardrails — PII redaction, injection protection, secret and keyword scanning, pre-call and post-call. Which ones run is resolved per request: the organization's guardrail switch first, then the narrowest applicable assignment wins across key > team > org > default, and a winner disabled at that scope does not run.
• Budget controls — spend limits per key, team and organization.
• Observability — cost and usage on billable calls. Free routes are genuinely free and carry no x-nr-request-cost: /v1/messages/count_tokens, and video polling and content retrieval.

Smart Router aliases and fallback chains carry two conditions worth knowing before you rely on failover you have not enabled:

• Opt-in by what you put in model. An alias gets the strategy and its chain; a concrete model is never re-routed and inherits no hidden fallback.
• Text wires only — chat completions, responses, messages and legacy completions. Audio, image and video calls take a single-provider route and are not cross-provider Smart Router wires.
• API reference — the wire contract every SDK here implements.