Launching today

SoreQen Code
The coding agent that checks its own work
34 followers
The coding agent that checks its own work
34 followers
SoreQen Code is an autonomous coding agent by ZorQelis AI. Describe a task and it reads your project, makes a plan, edits files, runs commands and tests, checks its own work, and tells you plainly what it did. Use it in the SoreQen desktop app or the open-source CLI, and approve steps from your phone when you are away. It runs on 18 models with one credit balance, and SoreQen Auto picks the right one for each step. Free to start.





Free
Launch Team

AWS Activate CreditsCredits that fuel your startup’s next milestone
Promoted
Hi Product Hunt, I am Ved, founder of ZorQelis AI.
SoreQen Code is the coding agent I wanted for myself: one I could hand a real task, leave running, and trust when it said it was finished.
How it works: you describe the task, and it reads the codebase, writes a plan, edits files, runs commands and tests, and checks the result. Every turn ends with a plain report of what changed, what passed, and what it could not verify.
A few things we cared about:
Control. Plan, Manual, Auto or Bypass. Auto checkpoints your project before it acts and asks before anything it cannot undo, like a push. Any turn can be rewound.
Goals. Give it an objective, and it keeps going turn after turn until the objective is met and verified.
Anywhere. The SoreQen desktop app (Windows and Linux; Mac in the next release), an open-source CLI under Apache-2.0, and your phone as the remote.
Models. 18 models on one SoreQen credit balance, with SoreQen Auto picking per step, 30 more through Flex, or your own API key.
It is free to start, with plans from $8 a month. I will be here all day, and I would love to hear what you would hand to it first.
Approving steps from my phone while the agent keeps running on the desktop is the one thing I have wanted on a long build, since I usually notice it is stuck an hour after I left. Does it still run the tests on the same 18 models, or does it fall back to the cheap one for those?
@karimbenkeroum Exactly that scenario is why we built phone approvals. Wake up an hour into a stuck build? Not anymore. And yes, tests still run across the same 18 models. There’s no silent fallback to a cheaper one, so a pass means the same thing it does on desktop. Would love to hear how it holds up on your next long build!
The checkpoint-before-it-acts part is what would actually get me to trust an agent with a real codebase — most agents I've tried will happily keep going after a bad edit and only tell you at the end. Curious how it handles a failed verification: does it retry the same step with a different approach, or stop and show you what it couldn't verify? The rewind-any-turn idea is the other piece that makes this feel usable day to day.
@jcbuilds Thanks! That's exactly the problem I wanted to address with SoreQen Code.
The important part is that it doesn't treat an unverified result as a successful one. Before reporting back, it checks the turn record. If a test wasn't actually run, it won't claim that it passed. The final report clearly separates what changed, what passed, and what couldn't be verified.
With /goal, it can continue working across turns toward the objective and verify progress along the way. If something remains unverified, it reports that honestly rather than pretending the task is complete.
And yes, /rewind is there for exactly those moments when an iteration doesn't go as expected. You can restore a previous checkpoint instead of having to manually undo everything.
I'm still interested in improving how the agent handles failed verification, so this is really useful feedback. Thanks for bringing it up!