Build modes
IntermediateMost AI builders give you one agent, one shot. bychat lets you change how the work gets done. Run a single agent for speed, run two in parallel and keep the better result, or — the one we're proudest of — have two agents work independently and a third review both and write one fresh, final answer. Same chat, different horsepower.
Review of two, in one picture
Agent A builds it its way
Agent B builds it independently
A third agent reviews both, then writes one fresh, final result
Two takes, one reviewer. You don't pick a winner — the reviewer reads both attempts, keeps what each got right, and produces a clean third version.
Single agent
One agent, straight through. The fastest, cheapest way to build — perfect for small changes and quick iterations where you just want the result.
Two in parallel
Two agents tackle the same prompt independently at the same time. You get two genuinely different attempts to compare instead of one — great when there's more than one good way to do something.
Review of two
FlagshipTwo agents work independently, then a third — fresh eyes, no ego in either attempt — reviews both, catches what each missed, and writes a single final result that's better than either input. It's a second opinion and a tie-breaker built into the build itself.
Branches
Spin the same idea into separate branches and explore directions side by side without overwriting each other. Keep the branch you love, drop the rest.
Post-run check
After any run, a check agent re-reads what was built against your prompt and the project's rules — flagging mistakes, regressions, and security issues before they reach your preview.
Pick the mode that fits the moment: fast when you're moving, review-of-two when it has to be right. And every mode can be followed by a check pass — which is where we go next.
How checks & security work