HumanLayer

HumanLayer is a multiplayer coding agent IDE and cloud that puts agent sessions, plan artifacts and diffs in one shared workspace, with structured workflows and comment-driven design reviews for engineering teams.

HumanLayer is a coding agent IDE and cloud built for teams that run AI coding agents at scale. Instead of leaving each engineer alone with a terminal session, it gathers agent sessions, plan artifacts and code diffs into one shared workspace where several people can follow the work, comment on it and redirect it. The product comes from the group behind the "context engineering" and 12 Factor Agents material, and its design reflects that background. The unit of work is a task. A task groups agent sessions, artifacts and git worktrees, and can be created inline or pulled from a Linear or Jira ticket. When setting one up, the user picks a directory, a workflow (such as RPI, RPI Outline, PRD-Oriented, Oneshot or Freeform), the repositories involved and the agent to use. HumanLayer then provisions the worktrees and starts the first session. Its flagship workflow, QRSPI, moves through six phases: questions, research, design, structure, plan and implement. The aim is to make the agent expose its misunderstandings early, so engineers keep control of architecture and code quality. Design documents are the main collaboration surface. Teammates and agents leave inline comments on the same document, and those comments feed back to the agents, so the document works as the interface to the code. Research notes, designs, plans and mockups are versioned and tied to the task. Sessions stay fully inspectable, down to individual thinking messages and subagent tool calls. A local daemon runs sessions in parallel on a laptop, cloud daemons take long-running work, and everything syncs through the HumanLayer API, with the same interface on web, desktop or phone. The tool uses a bring-your-own-key model for Claude Code, Codex, Copilot and Fireworks, so there is no separate token bill. It suits engineering teams working in large codebases who want a structured, opinionated process. Compared with plain terminal agents or generic doc tools, it trades flexibility for shared visibility and a defined workflow.