The boom in AI coding tools has opened a new race: not just for systems that write code, but for systems that try to organize everything around it. That is the space pm-workspace is aiming at. The public GitHub repository presents itself as a project management system powered by Claude Code, designed for CEOs, CTOs, project managers, scrum masters, architects, developers, and QA teams, with the ability to delegate technical implementation to AI agents.
The project is built around an assistant called Savia, described in the repository as the layer that keeps work flowing: managing sprints, breaking down backlog, coordinating coding agents, handling billing, generating executive reports, and tracking technical debt from within Claude Code. At the time of checking, the repository is public on GitHub, licensed under MIT, and shows 17 stars, 1 fork, and 603 commits. Those are still early open-source numbers, but they suggest this is more than a rough concept.
What makes pm-workspace interesting is that it is not trying to be a new model or a new IDE. Instead, it is building a workflow layer on top of Claude Code. The README organizes the experience by role, with separate quick starts for PMs, tech leads, developers, QA, product owners, and executives. It also describes a connected flow where time entries feed costs, costs feed invoices, invoices feed executive reports, and sprint items, specs, implementation, code review, tests, and DORA metrics all connect inside the same system.
That positioning reflects a real shift in how AI is being used in software teams. Writing code faster is only part of the value. In many organizations, the harder problem is coordination: turning backlog into clear specs, deciding what enters a sprint, preserving context between sessions, and connecting delivery work to reporting and governance. PM-Workspace is trying to solve that wider orchestration problem through a structured .claude setup made up of commands, agents, skills, hooks, and rules.
The repository layout reinforces that ambition. Under .claude, the project includes directories for commands, agents, skills, hooks, and rules. Around that core, it adds documentation, quick starts, scenario guides, scripts, tests, and a mobile companion project called Savia Mobile. In other words, the idea is not just to chat with Claude, but to operate a full project workspace around it.
One of the most notable angles is its emphasis on Spec-Driven Development. The README says tasks can be turned into executable specs, then handed off to agents that generate handlers, repositories, and tests across multiple languages while working in isolated worktrees. That fits a broader trend in AI-assisted development: moving effort away from “please write code” toward “define the work clearly enough for agents to implement it with less ambiguity.” PM-Workspace tries to formalize that process.
The scope claimed by the project is wide. The README references 396+ commands, 34 specialized agents, 41 skills, and 16 hooks, along with role-specific guides and sector-specific workflows. It also says the system covers sprint management, Monte Carlo completion forecasting, architecture pattern detection, technical debt prioritization, diagram generation, regulatory compliance, multi-cloud infrastructure support, persistent memory, executive reporting, accessibility support, and industry-specific command sets. Those claims come from the repository itself, so they should be read as the project’s stated capabilities rather than as an independent validation that every area is equally mature today.
Another practical detail is its integration story. PM-Workspace says it can work with Azure DevOps, Jira, or a fully Git-native workflow called Savia Flow. That gives it a broader operational pitch than many AI agent projects that only target individual developers. It suggests the system is trying to fit into real delivery environments rather than positioning itself as a standalone experiment.
The mobile angle also stands out. The repository includes an Android app project, Savia Mobile, described as a native Kotlin/Compose app connected to pm-workspace through Savia Bridge, an HTTPS/SSE wrapper around Claude Code CLI. The goal is to extend the same management and interaction layer beyond the terminal and desktop into a mobile interface with streaming chat and encrypted local persistence.
PM-Workspace also tries to present itself as a controlled environment rather than just an automation playground. Its README lists rules such as never hardcoding secrets, confirming before writing into Azure DevOps, never launching an agent without an approved spec, never running production Terraform without approval, and always working through branches and pull requests. That matters because one of the big concerns around AI orchestration is not what the system can do, but what guardrails exist when it starts doing more.
More broadly, the project says something about where the agent ecosystem is heading. The next layer of competition is no longer only about which model writes the best function. It is about which system can coordinate backlog, context, testing, reporting, governance, and delivery across a team. PM-Workspace is still early in public traction, but it is clearly aiming at that larger category: not the coding copilot, but the AI-powered project workspace built around it.