Aura Autopilot is an open-source AI playtester for Unreal, Unity and Godot

Ramen VR, the company behind the Aura game-development assistant, has published Autopilot, which it calls a “continually learning real-time general playtesting system”. The repository was created on GitHub on 30 October and is Apache-2.0.

What it does

You install Autopilot as a plugin and give it a prompt. The README lists four kinds:

• Exploration: “Explore the crafting system in Lvl_Main.”
• Testing: “Go through the currently open level and tell me everything that looks wrong.”
• Verification: “Verify that the tree shakes when you hit it with an axe.”
• Goal: “Kill the boss while keeping your health above 20% of its maximum value.”

For each prompt it launches a play-in-editor window, reads the project’s source code and starts playing. The README says it plays by having an LLM coding agent write harnesses for a decision model, based on what it knows about your game. It builds a project memory as it goes; Ramen VR warns that the first few runs will be “suboptimal” and says it “visibly improves the more it plays your game”.

Where it runs

• Engines: Unreal Engine 5.4 to 5.8, Unity 6 or later, Godot 4.7 or later, on Windows and macOS. Building from source is Unreal-only, through a PowerShell script.
• Accounts: the editor plugin uses an Aura plan or your own API keys. Autopilot also runs inside Aura for Unreal and Unity.
• Coding agents: autopilot connect claude or autopilot connect codex installs skills and an MCP server so Claude Code or Codex can drive it on your own subscription.

Setup asks whether to share your runs with Ramen VR to improve Autopilot. When you uninstall, it keeps your project’s memories unless you choose to delete them. The README does not say where the decision model runs or what a shared run contains, so check that before enabling sharing on an unreleased game.

Game engines have been moving the same way: Unreal 5.8 documents an experimental MCP server in the editor, and Epic’s AI MegaJam put agents to work on real builds.

Why this matters

Manual playtesting is slow, and scripted tests break whenever a level changes. Autopilot is an open, engine-neutral attempt to let an agent play the build from a plain-language goal, and it plugs into Claude Code and Codex. It is new and unproven, but the code is public, so studios can inspect what it sends before trusting it.