Rouvr

Rouvr is an autonomous engineering colleague that turns a conversation into tracked, independently verified work. You chat with Rouvr like a teammate. Every turn goes through an intake step that sorts it into chat, knowledge or a task. Plans and decisions are saved automatically to Knowledge, which is versioned project memory. When you describe a plan, Rouvr asks clarifying questions, saves the plan, and offers to Convert, Edit or Keep it. When you approve a plan, Rouvr turns it into a slice of tasks with contracts and pass criteria. It also writes hidden test suites that the coder never sees, each sealed with a timed receipt. Execute dispatches the tasks to a coding worker that claims them under a lease. Failure is part of the design. In our demo, a task's first attempt fails its criteria. Rouvr records the failure, reopens the task, reclaims it and gets a passing attempt. The task then moves through review to done, with every step kept as evidence. Each task shows a pipeline (Contract, Knowledge, Build, Review, Evidence) and a delivery timeline, so you can see what was agreed, what knowledge was applied and why the work counts as done. The stack is a FastAPI ledger (WorkHub-lite on SQLite), a Next.js workspace (Board, Chat/Plan, Knowledge, Escalations, Projects) and an MCP server, so other agents can use the same ledger. Rouvr's core was built during the hackathon, Oct 27 to Nov 4, 2026. We used Rouvr to track its own build. Known limits, stated honestly: review is a step in the lifecycle, not yet a separate reviewer agent. Sealing fails closed, so if a suite can't be written, the task stays pending rather than being marked sealed.