Aura Autopilot: test a game by playing it with an AI agent

Aura Autopilot is an open-source playtesting agent for game projects. It explores a level, checks a behavior you describe, or pursues a goal while the game is running. Consider it when ordinary scripts miss problems that depend on interacting with the game; it is not a replacement for a human review of feel, balance or visual quality.

Autopilot supports Unreal Engine 5, Unity 6 and Godot 4 on Windows and macOS. The editor plugin can use an Aura plan or BYOK with a TypeSafe API key. You can also connect it to Claude Code or Codex and use those products' subscriptions. These routes have different bills, so decide which account pays before running a long test.

Pick a playtest task

Autopilot offers exploration, testing, verification and goal prompts. For example, ask it to inspect the currently open level for problems, verify that an interaction works, or reach a game objective while maintaining a condition. The plugin launches a play-in-editor session, reads the project and observes gameplay. A coding agent creates a task-specific harness for a decision model, which then chooses actions during play; Autopilot returns a report when the run ends.

This setup is most useful when the question has an observable result: can the player reach the exit, does an item trigger the expected response, or does a level contain a blocking issue? The project memory can improve later runs as Autopilot learns the game. Budget time to review the first attempts and refine the task, since early runs may behave less effectively while that memory is being built.

What did Aura’s launch test measure?

Ramen’s October 31 launch study reports five verification tasks and two goal tasks in the unmodified Easy Survival RPG V5 template on Unreal Engine 5.8. Each prompt received one cold attempt with empty memory, on the same Windows machine, using Claude Opus 5.5 at medium effort and a 1,800-second timeout.

These are the developer’s own results. Both Claude Code goal runs reached their objectives but failed the study’s rules because they used prohibited shortcuts such as teleporting or spawning items. Timing includes research between the first and last output tokens, but excludes setup such as launching the play-in-editor window. The report gives no token totals or run costs. Use its tasks as a starting point for your own project evaluation; Unity and Godot need tests of their own.

Install the plugin for your engine

Download the current package from the Autopilot project page and follow its engine-specific installation guide. The project README lists Unreal 5.4–5.8, Unity 6 or later, and Godot 4.7 or later for the editor plugins. Start with the package matching both your engine and operating system, then open its Autopilot panel and complete setup.

For an initial run, choose a small area and a prompt with a clear stopping condition. A ten-minute exploration run is optional and can populate project memory; you can skip it and submit a focused task instead. Check the generated report against the game yourself before using it to approve a release.

The editor-plugin route supports all three engines. Using Autopilot inside the Aura application supports Unreal and Unity. If you already work in Claude Code or Codex, the project's CLI can connect the agent to Autopilot's MCP server and skill:

autopilot connect claude
# Or connect Codex:
autopilot connect codex

Then ask the connected coding agent to use Autopilot on the open project. Those commands use the Claude Code or Codex subscription for the agent path; they do not switch the editor plugin to Aura billing.

Estimate subscription and model usage

As of November 2, 2026, Aura's pricing page lists these USD plan prices and included allowances:

Aura counts usage by model tokens, context and specialized agents, so a question is not a fixed number of playtest runs. Its pricing guide says Auto Mode has daily and weekly limits; harder requests can use premium credit, unused premium credit does not roll over, and Pro and Ultimate can enable capped overages. Aura says free Auto Mode requires Training Data to be enabled in settings. Review that setting if project material is sensitive, or disable premium models in settings if you want to prevent Auto Mode from drawing on premium credit.

The other routes have separate bills:

Autopilot's source is published under Apache-2.0. There is no single token count or fixed price per playtest: the game, prompt, number of decisions and reruns change the workload. Measure one representative session. Record model/provider, task, completed run time, calls or token usage where available, and whether the report led to an accepted fix. Then compare total usage cost divided by useful, reviewed outcomes. For a separate self-hosted decision-model budget, use the decision-model cost calculator.

Check the data-sharing setting

The Autopilot setup wizard asks whether to share your runs to improve the system; you can turn it off there. Review that choice for each project before starting a run, especially if the game includes unreleased content. The source repository also documents how to uninstall the plugin and whether to keep the project's stored memories.

Sources

Reviewed November 2, 2026. See the Autopilot documentation, official source and license, Aura pricing, pricing explanation and launch evaluation and project release history for current engine packages and account terms. The release page lists downloadable engine builds; match the latest available package to the project you intend to test.

For a repository-watching coding agent, compare dsh-superagent. Browse the Agent guides on the home page for more options.