wtf is vehir-platform?

Vehir Platform is an experimental computing environment built specifically for AI agents rather than for human users. Most tools bolt an AI interface onto software that was designed for people first. Vehir instead designs every layer of the system, from its tools to its build process, so that an agent can use it directly and reliably. The project provides structured, discoverable input and output that an agent can read and act on without having to parse free-form text. Errors come with machine-readable remediation hints, and the system can describe its own capabilities to an agent rather than requiring a human-written manual. The source language used inside the platform, called IPM, is meant to be written and checked by machines, not edited by hand. Under the surface, Vehir combines several unusual pieces into one self-hosting system. It has its own small kernel that manages workers using shared memory, its own compiler that translates its source code directly into x86-64 machine code, and its own loader that runs the compiled results without depending on a typical application runtime. Files and compiled objects are stored by content, meaning identical pieces are automatically deduplicated and kept as immutable snapshots called generations. Changes are declared as a desired end state, and the system works out how to reach it. This is a Linux only project, needs a 64-bit processor, and works best with 4 GB of RAM or more. Installation is designed to be handed to an AI agent as an instruction rather than run manually, and the same applies to uninstalling it. The project describes itself as experimental, still under active development, and notes that all of its source code was written by AI. Anyone considering it should expect interfaces to change between releases.