Your engine
Unity, GameMaker or your own engine, through editor scripting and command-line builds.
Services Engineering
Coding agents inside your engine.
We connect agents to your editor, repository and build system, with the permissions and review rules your programmers set. They work on scenes, prefabs and gameplay code, and every change arrives as a pull request.
How it runs
One ticket through the setup.
- Handled by an agent
- Faster with AI
- Your team signs off
-
Ticket and criteria
A programmer or producer writes the task and what done looks like.
Your team signs off -
Reads the project
The agent loads your scenes, prefabs, conventions and the relevant code.
Handled by an agent -
Edits and scripts
It makes the change in the editor and the code, on its own branch.
Handled by an agent -
Plays and profiles
It runs the build, plays the scenario and records frame timings.
Handled by an agent -
Review and merge
Your programmers review the diff, the test run and the profile.
Your team signs off
Example run
What a day of it looks like.
An illustrative log of the setup at work. Your tools, numbers and review points will differ.
- Your team
Ticket: a moving platform between Platform_Start and Platform_Goal.
- Agent
Scene read: 214 objects, 3 matching prefabs, project conventions loaded.
- Agent
MovingPlatform script and prefab variant created on its own branch.
- Agent
Command-line build passed, 142 of 142 tests passed.
- Agent
Playthrough: platform reaches the goal in 3.2 s, frame time 6.1 ms.
- Agent
Pull request opened with the diff, test output and profile attached.
- Your team
Programmer review: one change requested, then merged.
What we set up with you
The pieces behind it.
Most of the effort goes into context and guardrails, so the agent knows your project and stays inside its lane.
- Editor connectionMCP servers that let agents query and edit the open project. We maintain gms-mcp, an open-source one for GameMaker.
- Project contextConventions, architecture notes and prefab schemas written where agents read them first.
- PermissionsBranch-only write access, no secrets in context, and commands on an allowlist.
- Builds and testsCommand-line builds, tests and scripted playthroughs the agent can run and read.
- Review flowPull requests with the diff, test output and profile attached, in your review tool.
- Cost trackingModel spend tracked per task and per team, with limits.
Where it fits in your studio
Your repository
Source control, branch protection and CI stay as they are.
Your reviewers
Only a programmer can merge.
Your code
Agents run on your accounts, and nothing trains on your code without agreement.
Starting out
A first pilot.
A first pilot picks one well-bounded job, such as UI wiring, tool scripts or bug fixes in one system.
From you
Access & a lead
Repository and build access, a sandbox machine or CI runner, and a programmer to review.
From us
A working setup
An agent setup your team keeps using, its guardrails documented, and a record of merged and rejected changes.
Next step
Tell us what you're making.
A few lines is enough. We'll reply within one to two business days with a suggested starting point.