| Auto-Switch routing | AI-judged complexity classification routes every prompt to the cheapest capable model across four tiers, with multi-model fallback per tier. |
| Universal auto keys | Keys whose alias starts with auto force auto-routing for every request regardless of the model name sent — drop-in for any OpenAI-compatible tool. |
| Control Panel | Set the model pool per tier and the classifier model from a web page; changes apply live with no restart. |
| 1,300+ models | Wildcard passthrough for 18+ providers, including the full OpenRouter catalog (343 models) and Amazon Bedrock. Add a provider key and its entire catalog goes live. |
| Usage Dashboard | Spend per day, per key, and per model with budget bars; block/unblock keys, set budgets, and delete keys from the dashboard. |
| AI Console | Chat workspace with preset task buttons and a live routing log showing tier, model, latency, tokens, and cost per request. |
| Lead Agent | On-demand agent: researches a company live (Perplexity), scores Cocolevio fit 0–100, writes LinkedIn + cold-email outreach, and creates the enriched company in Coco CRM. |
| Budgets & tracking | Per-key budgets, expiries, and full spend attribution; every request is logged with its cost. |
| Audio & embeddings passthrough | Whisper transcription and embeddings bypass auto-routing and reach their real models — voice and RAG flows work through the same key. |
| A2A agent registry | The Lead Agent is registered as an A2A-spec agent with a ▶ Run action in the native Agents page. |