Eris connects to the world's best AI models through OpenRouter, a unified gateway that provides access to every major provider through a single API key. Responses stream in word-by-word for instant feedback.
The picker always offers each provider's newest model: names are looked up from OpenRouter's live catalog, so they update on their own when a new version ships.
Auto (Best) Mode
Auto uses smart routing: coding and reasoning go to Claude, factual questions to Grok, research to Gemini, creative writing to GPT. Everything else goes to Claude first, with GPT last because it is slowest to start answering. If a model is unavailable, the next one in line takes over.
Frontier Models
Claude Opus 5.5Frontier Anthropic's flagship. Deep reasoning, coding, nuanced writing. 1M context.
GPT-6.1 Sol ProFrontier OpenAI's flagship. Versatile, strong at creative work. 1M context.
Gemini Pro LatestFrontier Google's newest Gemini Pro. Research and long documents. 1M context.
Grok 4.7Frontier xAI's flagship. Fast factual answers. 500K context.
Other
Sonar ProSearch — Perplexity, with built-in web search
Asks the Pandemonium Grid instead of a cloud model. Answers come from the grid's own documents and are written by Qwen 2.5 32B on the Devil's RTX 5090. Stays on until you pick another model; /grid or /g asks a single question.
A private AI infrastructure built for Revelex Corporation in Boca Raton, Florida. Every machine sits in one room — the CEO's office. No cloud dependency. No third-party access. No ceiling.
15 purpose-built nodes running inference, training, document ingestion, vector search, monitoring, building automation, and secure storage — all connected by a 10-gigabit backbone through an enterprise-grade Ubiquiti switch. Two NVIDIA RTX 5090s, an RTX 5080 and an RTX 2070 SUPER handle local AI workloads. A fleet of Apple Silicon machines powers the RAG pipeline. A 64-terabyte Synology NAS with enterprise RAID holds every model, every document, every backup.
Eris — The Intelligence Layer
Eris has access to seven frontier AI models — Claude Opus 5.5, GPT-6.1 Sol Pro, Gemini Pro, Grok 4.7, Sonar Pro, DeepSeek V4 Pro, and GLM 5.3 Prime — always the newest version of each, and in Auto she picks the right one for every question. Coding and reasoning go to Claude. Facts go to Grok. Research goes to Gemini. Creative work goes to GPT. She doesn't just answer — she routes to the sharpest brain for the job.
She searches the web in real-time. She has persistent memory — learns about you and remembers across conversations. She speaks aloud, generates images, finds restaurants, and streams every response word-by-word. She's not a chatbot waiting for instructions. She's a system that acts.
This is not a cloud subscription. This is infrastructure we own, models we command, memory we keep, and intelligence we trust.
