Everything Required to Scale Production AI
Nicrron abstracts provider fragmented APIs into a unified high-performance gateway with smart routing, instant caching, and strict zero-data retention.
Smart Auto-Routing (`nicrron/auto`)
Automatically route every prompt to optimal AI models. Utilize specialized virtual slugs like nicrron/reasoning (DeepSeek V4 Pro & o3-mini) and nicrron/coding (Claude Opus 5 & Kimi K2.7 Code) for maximum task precision.
Side-by-Side Model Arena
Benchmark LLMs concurrently on identical prompts. Compare real-time Time to First Token (TTFT), tokens/second throughput speed, response quality, and dynamic cost difference with LMSYS-style blind battles.
Organization Teams & Shared Wallets
Collaborate with engineering teams on pooled credit balances. Manage fine-grained Role-Based Access Control (Owner, Admin, Developer, Viewer) with secure teammate invitation links.
Spend & Reliability Webhook Alerts
Receive immediate Slack, Discord, or backend webhook notifications when credit balances cross low thresholds or daily token spend velocities spike, cryptographically signed with HMAC-SHA256.
OpenAI-Compatible Embeddings
Generate vector embeddings across OpenAI (text-embedding-3), Mistral (mistral-embed, codestral-embed), and Voyage (voyage-4 family) with multi-provider fallback chains, dimension truncation, and zero payload retention.
Prompt Caching Discounts (Up to 90% Off)
Automatically pass through upstream prompt cache discounts. Enjoy 90% discount on Anthropic cache reads and 50% discount on OpenAI cached prompt tokens with full telemetry auditability.
Live EWMA Health & Circuit Breakers
Sub-millisecond routing telemetry tracks Exponentially Weighted Moving Average (EWMA) latency across providers and automatically trips circuit breakers on degraded endpoints to prevent downtime.
Telemetry Logs & CSV/JSON Export
Inspect real-time API traffic filtered by date range, provider, status code, and cost. Export full audit logs directly to RFC 4180 CSV for Excel accounting or structured JSON for observability pipelines.
Zero Data Retention (ZDR) Privacy
Security and privacy without compromise. Nicrron streams LLM prompts and embeddings strictly in-memory (RAM) and never stores prompt contents or completions to disk or database.