Application Ecosystem

What teams build on nRouter.
Architectures in production.

From autonomous code assistants to real-time voice agents and high-assurance financial screeners. Explore reference architectures running on nRouter with unified keys, built-in guardrails, and sub-millisecond routing compute.

100% OpenAI API compatible · zero lock-in · one monthly invoice

  • 153+Live ModelsAlibaba US, OpenAI, AWS Bedrock, Azure Foundry, Google Vertex AI & Anthropic
  • 1 API KeyUniversal AccessAll providers unified
  • < 1msRouting ComputeInternal dispatch step
  • 100%OpenAI CompatibleZero code rewrite
Reference Architectures

Proven patterns for production AI workloads.

Every pattern runs on the identical standard API. Switch models, activate guardrails, and enforce budget limits without touching application logic.

Developer Tools
REF-coding-assistant

Autonomous Code & Refactor Assistant

Sub-second TTFT with tool-use validation and failover

High-throughput coding agent that routes fast completion queries to cost-efficient models while escalating complex multi-file refactors to high-reasoning flagships.

< 240ms
Avg TTFT
4.2x savings
Token Efficiency
99.99%
Uptime
Model Mix:
DeepSeek-V4 ProClaude Sonnet 4.5GPT-5
Guardrails:Secret scanning · Prompt injection detection · AST validation
const completion = await openai.chat.completions.create({
  model: 'deepseek/deepseek-chat',
  messages: [{ role: 'user', content: 'Refactor this auth middleware...' }],
  extra_body: {
    fallback_models: ['anthropic/claude-sonnet-4-5'],
    guardrails: ['secrets-scanning', 'prompt-injection'],
  },
});
Conversational AI
REF-support-copilot

Enterprise Customer Support Copilot

Strict PII redaction and per-seat budget enforcement

Multi-turn customer support agent handling live customer chats, ticket triaging, and CRM updates with inline PII scrubbing before prompts reach model providers.

100% inline
PII Scrub Rate
< 350ms
Response Target
$0.02 / ticket
Budget Cap
Model Mix:
GPT-5 MiniClaude Haiku 4.5Gemini 2.5 Flash
Guardrails:PII redaction · Brand tone consistency · Toxic content filter
const reply = await openai.chat.completions.create({
  model: 'openai/gpt-5-mini',
  messages: ticketHistory,
  extra_body: {
    guardrails: ['pii-redaction', 'sentiment-safety'],
    budget_scope: 'team_support_tier1',
  },
});
Knowledge & RAG
REF-multilingual-rag

Enterprise Knowledge RAG Engine

Long-context document synthesis with grounded citations

Knowledge retrieval pipeline indexing millions of enterprise documents, synthesizing context-grounded answers across 40+ languages with automatic hallucination checks.

Up to 200k
Context Window
38%
Cache Hit Rate
42 supported
Languages
Model Mix:
Cohere Embed v3Llama 3.3 70BClaude Sonnet 4.5
Guardrails:Hallucination check · Prompt leak protection · Source citation
const answer = await openai.chat.completions.create({
  model: 'meta/llama-3.3-70b-instruct',
  messages: [{ role: 'user', content: formattedQueryWithChunks }],
  extra_body: {
    cache_policy: 'semantic_match_0.92',
    temperature: 0.1,
  },
});
Document Processing
REF-document-intelligence

Contract & Financial Document Analyzer

1M context extraction with schema-validated structured output

Automated financial and legal document analysis parsing 500+ page filings into strict JSON schemas with complete audit logging and cryptographic verification.

1M tokens
Max Document
99.98%
Schema Match
SOC 2 CC7.2
Audit Trail
Model Mix:
Gemini 2.5 ProClaude Sonnet 4.5
Guardrails:Schema validation · Strict JSON enforcement · Append-only audit
const analysis = await openai.chat.completions.create({
  model: 'google/gemini-2.5-pro',
  messages: [{ role: 'user', content: contractMdx }],
  response_format: { type: 'json_object' },
  extra_body: { audit_tag: 'legal_matter_8820' },
});
Realtime Voice AI
REF-voice-agent

Sub-150ms Conversational Voice Agent

Bi-directional audio streaming with real-time token tracking

Inbound and outbound telephone agents conducting natural voice conversations with sub-150ms time-to-audio, interruption detection, and minute-by-minute credit budgeting.

< 140ms
Audio Latency
Instant
Interruption
Sub-second meter
Billing
Model Mix:
OpenAI Realtime APICartesia SonicWhisper v3
Guardrails:Audio redaction · Interruption handling · Max session duration
// Connect WebSocket to nRouter Realtime gateway
const ws = new WebSocket(
  'wss://api.nrouter.ai/v1/realtime?model=gpt-5-realtime-preview',
  ['realtime', `nrouter-key.${process.env.NROUTER_KEY}`]
);
FinTech & Security
REF-fintech-screener

Multi-Model Fraud & Compliance Screener

Dual-model arbitration with zero-content logging policy

High-security financial transaction screener running dual-model parallel evaluation to flag anomalies without ever storing payload bodies in long-term logs.

Zero content
Log Retention
99.4%
Consensus Rate
< 180ms
Eval Speed
Model Mix:
Claude Sonnet 4.5GPT-5Mistral Large 2
Guardrails:Zero-content policy · Financial entity redaction · FIPS compliance
const screen = await openai.chat.completions.create({
  model: 'anthropic/claude-sonnet-4-5',
  messages: transactionPayload,
  extra_body: {
    consensus_eval_with: 'openai/gpt-5',
    data_policy: 'zero_content_retention',
  },
});
Submit Your Architecture

Feature your production AI build.

We curate builds operating with real production traffic on nRouter. Submissions are reviewed weekly and published with complete architecture notes and latency proofs.

01

Verify Production Traffic

Ensure your application is active on nRouter with consistent traffic and configured virtual keys.

02

Document Architecture

Provide an architecture overview: models used, routing rationale, guardrails, and latency targets.

03

Editorial Review

Our engineering team reviews your build and features it in the showcase with code and benchmark proofs.

Build without limits

Deploy your next AI application on nRouter.

One OpenAI-compatible endpoint. Access all leading models, enforce budgets, and protect prompts with inline guardrails.

Production-grade · transparently priced · nRouter