Multi-Agent Development Orchestrator
Takes a request in plain language and decomposes it into a chain of ten specialised coding agents — architect, backend, mobile, QA, security, review, localisation — each running as an isolated process against a target repository.
A lightweight router model picks the chain for a given request. The agents never talk to each other directly; they coordinate through a shared specification and task state, which is what keeps a ten-step run from turning into a game of telephone.
If the router model is unavailable, deterministic keyword routing takes over. The component that chooses the chain is the obvious single point of failure, so it is the one part with a dumb, predictable backup.
Output streams over Server-Sent Events while the run is still happening. On protected repositories it branches automatically, and every run ends with a diff, a commit-or-revert decision and a cost estimate.
- Deterministic keyword routing as a fallback when the router model is unreachable.
- A run that stops dead because the one component choosing the chain went down.
- Ten specialised agents, isolated per process
- Router model with deterministic keyword fallback
- Live output over Server-Sent Events
- Auto-branching; every run ends in a decision