Whip

WHIP

// A Dual-Council Multi-Agent Development Platform

Describe what you want.
Whip plans it, builds it, ships it.

Whip pairs a planning council that discovers what you actually need with an execution council of isolated, specialized coding agents — so you go from a one-line idea to a running, reviewed application without managing the pipeline yourself.

A run, replayed

working

It cannot merge your code

Whip opens a pull request and stops. There is no merge endpoint, no force-push and no branch deletion anywhere in the tree, and a test fails the build if one appears.

Every diff passes real checks

Syntax, write scope, compile and the project's own test suite, then a reviewer against the task's acceptance criteria. They fail closed: a check that cannot run is a rejection, never an approval.

It looks things up

Workers search the live web and every claim that survives is attributed to a source the search actually returned. An answer from memory is dropped rather than quietly repeated.

Council 1 · Planning

A LangGraph-powered discovery agent asks the handful of questions that matter, then turns your idea into a concrete implementation blueprint.

Council 2 · Execution

Native ReAct workers run specialized build, frontend, and database agents against your workspace, with mechanical gates and a reviewer on every diff.

Live everything

Execution status, stage, and terminal output stream over WebSockets in real time, from the first clarifying question to the final commit.

How a project comes together

  1. 01

    Name your project — that's the only input the New Project flow asks for.

  2. 02

    Council 1 opens a chat, asks a few clarifying questions, and finalizes a spec.

  3. 03

    Council 2's worker agents execute the plan against your isolated workspace.

  4. 04

    Watch execution progress and diffs stream live until it's done.