Introduction

Ara is a game engine written in Rust, with a native editor, C# 14 / .NET 10 gameplay scripting and built-in tools for AI agents. This manual shows you how to install it, how a project is put together, and how you and your agents can work on the same project without stepping on each other.

#What makes Ara different

Every edit is a typed command. The editor, the ara command-line tool and agents connected over MCP all change a project through one shared command registry. Each change becomes a transaction that records who made it and why, and you can undo it. See Commands and transactions.

Projects are plain text. Scenes and prefabs are canonical JSON: a flat map of records keyed by short IDs, with @path#id references between files. Because the format is stable, ara merge can merge two versions record by record, and git can use it as a merge driver. See Documents and IDs and Merging.

Agents are first-class users. A built-in MCP server gives Claude Code, Codex and other MCP clients the same commands the editor uses. Agent edits apply immediately, like a teammate's, and you can undo any agent turn. See Connect an agent.

The engine explains itself. Diagnostics carry stable codes such as ARA1302. Ask ara why name:Crate not visible and the engine answers with ranked causes, the evidence for each and fixes you can run as they are. See Diagnostics.

#What's in 0.0.2-alpha.3

Ara is in public alpha, for testers and agents, and this manual describes the 0.0.2-alpha.3 preview. It includes:

• Ara Hub, which installs engine versions side by side, creates projects from templates, opens them in the editor or headless, and gives you the setup for your agents.
• The editor, with hierarchy, inspector, assets and console panels, tool panels for agents, materials, VFX, animation and profiling, detachable windows, and Play in a separate, crash-isolated player process.
• The ara CLI and daemon: init, do, query, inspect, check, fmt, merge, why, logs and more. Run ara --help for the full list, or see the CLI reference.
• MCP for agents, either one global server through the Hub or one server per project.
• C# gameplay with live code reload, and standalone Windows exports with NativeAOT. See the Scripting overview.
• Engine systems: Jolt physics, spatial audio, navigation and crowds, animation with IK and motion matching, GPU VFX, terrain and foliage, world streaming, 2D, localization and saves.

#How this manual is organized

• Start gets you running: Install, Your first project and Connect an agent.
• Concepts explains the model underneath: Projects, Documents and IDs, Entities and components, Commands and transactions, Review and undo, Diagnostics and Merging.
• Agents covers the MCP server and Sessions and scopes.
• Scripting covers C# gameplay, starting at the Scripting overview.
• Reference is generated from the engine itself: every component and type, command, Engine Protocol method and diagnostic code. Your installed engine can generate the same pages for your project with ara docs --llms.

#Conventions in this manual

• Commands are shown as you would type them in PowerShell. ara accepts JSON arguments with or without quotes, so they work in other shells too.
• Output blocks come from real installs of the alpha. IDs such as e_qg8h and tx_n4r61zfz are minted per project, so yours will differ.
• Units follow the engine: meters, seconds, kilograms and degrees, with Y up.