Skip to main content

Core concepts

Before diving deep into p0, familiarize yourself with the key concepts that shape how you work with the AI coding assistant. These building blocks work together to create a safe, productive development workflow.

Foundational concepts

Workspaces

Workspaces

A workspace is a container for one or more git repositories, along with shared settings, coding standards, and plugins. Workspaces let you:
  • Work across multiple repos in a unified environment
  • Define project-wide conventions and preferences
  • Share context between sessions
  • Install workspace-scoped plugins
Think of a workspace as your project’s home base in p0.

Sessions

Sessions

A session represents a single coding task within a workspace. Each session:
  • Gets its own isolated git worktree for safety
  • Has a dedicated chat thread for communication
  • Can run in Plan mode (read-only) or Agent mode (full execution)
  • Tracks all changes and tool executions
Sessions keep your work organized and your main branch protected.

Modes

Plan and Agent modes

Two modes for direct coding tasks within p0:Plan mode (read-only):
  • Analyzes your codebase without making changes
  • Generates detailed implementation plans
  • Researches files and dependencies
  • Safe for exploration and brainstorming
Agent mode (execution):
  • Implements changes using real tools
  • Runs bash commands, edits files, searches the web
  • Shows real-time tool output
  • Creates commits and manages git
Use these modes when you need hands-on exploration or direct coding outside the spec-driven workflow.

Spec-driven workflow

Spec-driven workflow

p0 follows a five-phase pipeline for complex features:
  1. Import: Upload your product spec and optional references
  2. Refine: Polish specs with AI assistance
  3. Watch: Monitor AI implementation in real time
  4. Review: Review changes and create pull requests
  5. Post-PR: Track PRs and wrap up
This structured approach ensures quality at every step.

Tools

Real-time tools

When running in Agent mode, p0 uses tools to interact with your codebase:
  • Bash: Run terminal commands
  • File operations: Read, write, edit files
  • Web search: Research APIs and documentation
  • Git operations: Commit, branch, diff
Every tool execution is visible in real-time with full output, giving you complete transparency into what p0 is doing.

Sub-chats

Sub-chats

Complex sessions can have multiple conversation threads called sub-chats:
  • Break large tasks into focused discussions
  • Work on parallel concerns simultaneously
  • Keep related context together
  • Switch between threads without losing progress
Sub-chats help manage complexity in long-running sessions.

How it all fits together

The primary way to use p0 is the spec-driven workflow. Here’s how a typical project flows:
  1. You create a workspace for your project and add repositories
  2. You describe a feature in Spec Mode — p0 asks clarifying questions and brainstorms with you
  3. p0 writes a detailed product spec for your review — this is a critical checkpoint before building
  4. Once you approve the spec, p0 builds your feature using real tools (bash, file edits, web search) in an isolated git worktree
  5. You use sub-chats to manage complexity across different aspects of the build
  6. p0 creates pull requests for your team to review
  7. You merge and ship
For exploration, quick fixes, or tasks that don’t need a full spec, Plan mode and Agent mode are available as direct coding tools within any session.

Git worktree isolation

One of p0’s most important safety features is git worktree isolation. Each session runs in its own worktree, which means:
  • Your main branch stays clean and untouched
  • Multiple sessions can run concurrently without conflicts
  • You can experiment freely without risk
  • Failed sessions don’t pollute your workspace
Learn more about git worktree isolation.

Available models

p0 is powered by Anthropic’s Claude models. You can switch between models using the /model command or the model selector in the chat header:
  • Claude Sonnet — fast, efficient performance for most coding tasks (default)
  • Claude Opus — maximum capability for complex reasoning and planning
  • Claude Haiku — lightweight model for quick tasks

Next steps

Now that you understand the core concepts, explore specific areas:

Workspaces

Learn how to create and configure workspaces

Sessions

Master Plan and Agent modes

Spec-driven workflow

Follow the five-phase pipeline for complex features

Tools

Understand how p0 interacts with your codebase