Everyone’s raising their AI lobster. Not everyone has a server to run it on.

Agenvoy is a pure Go agentic backend — no framework overhead, no bloated runtime. Verified running on a Raspberry Pi Zero 2W with just 512MB RAM. Multi-Provider LLM with Intelligent Routing Agenvoy integrates seven AI backends — GitHub Copilot, Claude, OpenAI, Gemini, Nvidia NIM, and any OpenAI-compatible endpoint (Compat/Ollama) — behind a unified Agent interface. A dedicated planner LLM automatically selects the most appropriate provider for each request, eliminating the need to manually switch models. Named compat[{name}] instances allow multiple local model endpoints to coexist, each with independent URL and credential configuration. Skill-Based Agentic Execution Skills are declarative Markdown files ( SKILL.md ) that define a task's system prompt and tool allowlist. At runtime, a Selector LLM picks the best matching skill across 9 standard scan paths, then drives a tool-call loop of up to 128 iterations until the task completes. When the iteration limit is reached, the engine auto