# EVOX — Mission Control

An agent orchestration system that gives AI engineering agents persistent state outside the model, hierarchical memory, a shared communication layer and a scheduler, so work continues across sessions instead of starting over each time.

EVOX is an agent orchestration system built around five stated constraints on how large-language-model agents work: they have no memory between sessions, so state must live in a persistent layer read on boot; context windows are finite, so memory must be layered into working state, daily notes and long-term memory; multiple agents on one codebase will collide without a shared communication layer; agents never wake up on their own without a scheduler or event system; and agent output is not fully reliable, so actions need permission levels separating what an agent can do alone from what needs human approval.

From these constraints, EVOX positions itself as a coordinating layer for an AI engineering team, keeping agents working continuously by externalizing memory, mediating communication between agents, scheduling their activity and enforcing permission boundaries.

It is written in TypeScript.

Website: https://github.com/sonpiaz/evox