Launching today
PerfLint
A senior tech lead, inside your Unity Editor.
3 followers
A senior tech lead, inside your Unity Editor.
3 followers
One scan finds what's slowing, bloating, or breaking your Unity project — performance, assets, migration, project settings. Detection is a deterministic rule engine, not an LLM guessing: same findings every run, offline, zero tokens. Autopilot ranks them by evidence and applies only what it's sure of, never anything that changes how your game plays. Static scan plus Play Mode profiling. Run it in the editor, from the CLI, or hand it to your AI agent over MCP. Local-first, zero telemetry.
Hi everyone,
PerfLint scans your Unity project and tells you what's slowing it down, bloating the build, or about to break on a version upgrade. Every finding comes with a severity, an exact file and line, and a fix.
This came out of my own projects. The bigger they got, the more redundant assets piled up, and the performance problems followed — stutter, dropped frames, memory creeping up until it mattered on device. None of it announces itself. You notice once the build is already too big or the frame rate is already gone, and then you're spending a day in the Profiler working backwards.
The decision I'd defend hardest is that detection is a plain rule engine, not an LLM. Same project, same findings, every run. The model only shows up afterwards, to explain something or write a fix snippet. I've used enough tools that answer differently each time you ask.
Autopilot is the part I'd most like people to poke at. Instead of a Fix All button it works one round at a time, applying only what the evidence is strongest for. Anything that would change how your game plays or looks, it leaves for you to decide. That makes it noticeably less aggressive than it could be, and I'm genuinely unsure whether I've drawn that line in the right place.
Here's the run that convinced me this was worth building. Unity's Viking Village sample: 852 findings, 204 auto-fixable. Before touching 95 model importers the agent stopped to check whether disabling Read/Write would break anything at runtime, found the one script writing to a procedurally built mesh, confirmed that one was unaffected, then applied all 204 with zero failures. Grade D to C in 1 minute 41. The Fix All button makes the same changes faster, but it can't do that check.
Free covers the full scan, every finding and a shareable HTML report. Pro is $9.99/mo; PHYEAR takes 50% off the first year, good through September 7.
Point it at your messiest project and tell me what it missed, or where it got in your way. I'd much rather hear that it broke on you than not hear it.