What is Piyaz

Most of us are not writing much code by hand anymore. We direct agents that do. An agent will plan and ship one well-scoped task reliably. A real project is dozens of them with dependencies between them, and that is where it gets hard: context drifts between sessions, attention lands on the wrong things, and quality falls off the further the work gets from the one slice the agent actually reasoned about.

Piyaz brings the conventional engineering process to agentic work. You drop in a project or a feature, and it decomposes the work into small, concrete tasks with the dependencies and relationships between them. You and your agents then refine those tasks, plan them, implement them, and ship them, without losing the thread between sessions or between agents.

The app is the memory and the coordination layer. It keeps a consistent record of the project, maps how the tasks relate, and orchestrates the work across people and agents. When an agent picks up a task, the context it needs is already attached: a token-dense brief engineered around how model attention works, so the agent arrives knowing the project, the engineering standards, and the decisions you made along the way in your codebase. It does not start cold, and it does not need a re-briefing.

That is the shape of it: an agentic workspace where people and AI coding agents share one understanding of every project.

Two core concepts

• Context network: a living map of your project that captures not just what was built, but why decisions were made, what was tried and abandoned, and how parts of the codebase relate.
• Context retrieval interface: five context shapes, one per job. A task reads as a quick summary, a working view for refining, a planning view for writing the implementation plan, a full agent view for coding, and a review view for checking finished work. Each stage gets the slice it needs, not the whole project.

Together they drive the full lifecycle: Brainstorm → Decompose → Refine → Plan → Execute → Track.

What you get

• Hosted at app.piyaz.ai: the default way to use Piyaz. Create an account, install the plugin, and sign in. It works across every project you open, with no server to run and no database to manage.
• A plugin for your coding agent: Claude Code, Codex, Cursor, and Antigravity. It carries the /piyaz skill and the workflow skills (brainstorm, decompose, manage, and more); on Claude Code it also ships dispatchable agents.
• Composer: on Claude Code, an end-to-end loop that researches a task against your codebase, writes the plan, implements it on a branch, opens a pull request, reviews it and fixes what review flags, then picks up the next ready task. It can also merge for you when you authorize it. See the composer pipeline.
• Six MCP tools (piyaz_project, piyaz_task, piyaz_edge, piyaz_query, piyaz_context, piyaz_analyze): the programmatic surface your agent calls directly.
• A web app: Structure mode for the task list and detailed workspace, Graph mode to observe the context network.
• Open source: Piyaz is AGPL-3.0. Run your own instance on PostgreSQL and Next.js whenever you want. See the self-host guide.

Who it's for

Engineers who work primarily through AI coding agents (Claude Code, Codex, Cursor, Antigravity) on projects big enough that context from one session has to reach the next agent, and the one after that.

---