Unified middleware between every agent and every LLM provider. Budget enforcement, failover, caching, priority queuing, and cost attribution — at the infrastructure layer.
Agent Request → Auth → Budget Check → Priority Queue → Cache Lookup → Model Alias → Provider Select → Circuit Breaker → Dispatch → Cost Ledger.
Per-agent monthly and daily spend caps. Hard stop on excess. CFO approval flow for budget raises. Forecasting with anomaly detection.
Per-provider-key failure tracking. 5 consecutive failures opens the circuit — automatic fallover to next key. Zero-downtime provider outages.
Three-tier concurrency: Interactive (user-facing), Background (agent tasks), Batch (bulk processing). Interactive never blocked by lower tiers.
Tier 1: exact SHA-256 match (sub-ms). Tier 2: pgvector cosine similarity (>0.95 threshold). Eliminates redundant LLM calls across your entire AI workforce.
"fast", "smart", "cheap", "vision", "code", "reason" resolve to cheapest available model in that capability tier. Auto-adapts to pricing changes.
1,000+ model pricing definitions. Per-request cost calculation in micro-USD. Raw audit trail + daily aggregates. Real-time dashboards via SSE.
14 supported providers: Anthropic, OpenAI, Google Gemini, AWS Bedrock, Azure OpenAI, AWS SageMaker, Groq, Ollama, LlamaSwap, OpenRouter, Together, Fireworks, DeepSeek, DashScope.
Native OAuth 2.0 provider flow for Claude Code, GitHub Copilot, and Google Vertex AI — PKCE support and automatic token refresh.
Multiple keys per provider with automatic rotation on failure. AES-256-GCM encrypted vault, prefix validation on entry.
Free, self-hosted, and working for you in minutes.
Qorven