Verified fix
The original code failed, and the fixed version then ran without errors.
AI code debugging with execution-based verification
Stop guessing. Run the fix. See the evidence.
APReason helps you investigate bugs, reproduce failures, execute proposed fixes in a sandbox, and compare the results with a clear code diff. Know what passed the test — and what still needs work.
The problem with untested AI fixes
A suggested fix can look right and still fail the moment you run it. Without running the code, an assistant is predicting what will happen. APReason tests the fix against your actual failure and shows you the result, including when it doesn't pass.
Predicted fix
Code is suggested. You find out if it works when you run it yourself.
Execution-checked fix
The failure is reproduced, the fix is run, and the outcome is shown with a diff.
How the debugging workflow works
- 01
Reproduce
Your original code runs exactly as given, so the real error and traceback are captured instead of guessed.
- 02
Investigate
APReason reads the real failure and explains the likely causes before proposing a change.
- 03
Run the fix
The proposed fix runs in the same sandbox, against the same reproduction.
- 04
Show the evidence
You get the real error, a before → after diff, and the output of the fixed run.
A real example
This Python snippet and its output are reproducible: run it yourself and you will get the same error. The layout mirrors the verification card APReason shows above its explanation; file paths are shortened.
def average(nums):
return sum(nums) / len(nums)
print(average([]))Traceback (most recent call last):
File "main.py", line 4, in <module>
print(average([]))
File "main.py", line 2, in average
return sum(nums) / len(nums)
ZeroDivisionError: division by zero def average(nums):+ if not nums:+ return 0.0 return sum(nums) / len(nums) print(average([]))0.0Verified here means the empty-list case no longer crashes. Whether returning 0.0 is the right behaviour for your program is still your call.
Every result is labelled honestly
Unverified fix
The fixed version still failed. You see that failure, not a confident claim.
No failure reproduced
The original ran cleanly, so there was nothing to verify against. APReason says so.
Runner problem
If the sandbox itself fails, it is reported as a runner issue, never as a bug in your code.
Test one hypothesis without losing the others
Debugging rarely follows one line. Open a branch from any part of an answer to dig into a single suspected cause, then return to the main investigation where you left it. Learn how branching works, or see every hypothesis on a visual thought map.
Runtimes and limits
Supported
- • Python
- • JavaScript (Node.js)
- • TypeScript (Node.js)
- • Up to 8 extra files per run
Not a fit today
- • Compiled languages (C++, Rust, Go, Java)
- • Long-running jobs — JS/TS runs stop after 20 seconds
- • Code that depends on your own servers, databases or secrets
Debugging sessions per day
One session covers a whole answer's run-and-verify attempts.
- Free · $0/mo
- 1
- Pro · $7/mo
- 25
- Premium · $18/mo
- 50
Frequently asked questions
Does APReason actually run my code?
Yes, for supported languages. When you share broken code, APReason runs it in a hosted sandbox to capture the real error, then runs the proposed fix.
What does “verified” mean?
It means the original code failed and the fixed code then ran without errors, under the tested conditions. It does not prove the fix is correct for every input or that all intended behaviour is preserved.
Which languages are supported?
Python, JavaScript and TypeScript (run with Node.js). Compiled languages such as C++, Rust and Go are not executed today.
What if my code runs for a long time?
JavaScript and TypeScript runs stop after 20 seconds. Long-running jobs, servers and code that needs live outside services are not a good fit for the sandbox.
Can I include more than one file?
Yes. Up to 8 extra files (such as helper modules or sample data) can be placed next to the code before it runs, within your plan's context limit.
What happens to the code I share?
Code you share is sent to our AI and sandbox providers to be analysed and run, and is stored with your conversation. See our Privacy Policy for details. Privacy Policy.
Can I try it for free?
Yes. Free accounts get 1 debugging session per day. One session covers a whole answer's run-and-verify attempts.