A Spec-Driven Multi-Agent Framework for AI-Assisted Software Engineering
APM is an open-source framework for managing ambitious software projects with AI assistants. On long tasks, the LLM behind your agent suffers from and as its conversation grows. APM counters the first by recording the project's intent in planning documents, and the second by splitting the work among agents in separate conversations. Coordination between the agents is mediated by you, making each exchange a chance to review the work and intervene before issues propagate. We call this .
An APM session runs in two phases. The writes three planning documents with you: the Spec, the Plan, and the Rules. The then hands out assignments and carry them out. Their messages pass through , and the lives in shared files outside any conversation. Any agent can still offload work to its own subagents for efficiency.
The documentation covers the workflow and the framework's design in full.
APM supports Claude Code, Codex CLI, Cursor, GitHub Copilot, Antigravity, and OpenCode. It installs through agentic-pm, a command-line tool on npm that needs Node.js 18 or later:
npm install -g agentic-pm
In your project directory, run the command below to install APM's commands, guides, skills, and document templates. Select your assistant when the CLI prompts you, or pass it with -a:
apm init
Open your assistant in the project and run:
/apm-1-initiate-planner
Describe what you want to build or change. The Planner works with you on the planning documents, then directs you to start the Manager in a new conversation. From there, each agent tells you what to run next and where. For the full walkthrough, see Getting Started.
APM runs inside your AI assistant. A command-line tool, agentic-pm, installs its files into your project directory from your terminal.
For installation and first steps, see Getting Started.