Launching today
ByteAsk Cloud Agents
Find the fix in big codebases. Ship it from Slack.
53 followers
Find the fix in big codebases. Ship it from Slack.
53 followers
ByteAsk gives Claude Code and Codex an index of your whole repo, so they start in the right file. Its Cloud Agent takes bugs from Slack, Jira or the browser, builds and tests in a sandbox, and opens a draft PR. Start with $50 of credit, no card.
Free Options
Launch Team
AWS Startup AdvisorPersonalized AWS guidance that meets you where you’re already building
Promoted
Hey Product Hunt 👋 Anirudha and Pratyush here (ex-Optiver, ex-ThirdAI, YC F26).
> SOTA on Multi-SWE-bench (C/C++)
> 60% fewer tokens than Claude Code and Codex
> Auto debug / perf / optimize / test with native C/C++ tools
> Free plan. Open-source CLI: github.com/ByteAsk/cli
Why we built it: in C++ the name isn't the code. Ask an agent to fix validate(order) and it greps, finds three functions called validate, reads all of them and edits the wrong one. The wrong edit still compiles. Which overload runs, which template gets instantiated and which #ifdef branch is live all depend on your build, and text search never sees the build.
What ByteAsk does:
Context indexes your repo the way your build does (LLVM in 13.7 min; Graphify hadn't finished after 3 h). Connect it with one prompt in Claude Code or Codex.
The harness builds with your toolchain, runs only the tests your change reaches, and uses sanitizers, gdb and perf on failures.
Cloud agents take bugs from Slack, Jira or the CLI, try fixes in parallel and send one PR with the proof attached.
Honest numbers (papers, code and tasks at byteask.ai/research): against Codex CLI on the same OpenAI model, both fixed 129 of 144 runs. ByteAsk didn't fix more, but it used 63–80% fewer tokens and cost 59–70% less. The biggest savings were on little-known projects the model hasn't memorized.
Free plan: $5 of usage a month, one cloud agent task, no card. On-prem or air-gapped? Book a call.
Tell us where it breaks on your codebase. That's the feedback we want most.
congrats on the launch anirudha. to your question, where i would expect it to break is monorepos with shared packages. agent fixes one app and quietly breaks another app that uses the same package. does it run tests for all affected packages before the pr, or only the one it touched?
congrats on the launch -- looks super cool!!
The 'agent greps around and lands in the wrong overload' problem is my daily life — I run Claude Code on my own repos all day and the first file it picks is wrong more often than I'd like to admit. The index updating on push plus picking up uncommitted changes is the part that would actually make me trust it. Curious how it handles Swift, though — the benchmarks and examples are all C/C++, and my world is Swift packages where the build graph looks pretty different.
the validate(order) example is painfully real, three functions with the same name and the agent picks the one that still compiles
also like that you posted same fixes as Codex, just cheaper, instead of a made up win rate. does the index rebuild on eveyr commit or only when the build config changes??? on a big repo that feels like where the time goes
@thamibenjelloun yes, it updates automatically whenever you push to main, but it doesn't reprocess the entire codebase. It only updates the files that changed and keeps everything else from the previous version.
It also doesn't need to recheck your build configuration unless something changes there, like compiler flags or the toolchain.
Even your uncommitted local changes are picked up, so the agent always works with the code you're actually looking at, not just what's on main.