Phasr is an open-source AI coding agent workspace designed to enable parallel development workflows. By orchestrating multiple AI agents simultaneously, it allows engineering teams to significantly increase their development throughput. The platform solves common orchestration challenges by providing a dedicated workspace that handles task distribution, environmental isolation, and a structured human-in-the-loop review pipeline. It is built to support various AI models and CLI-based coding agents, ensuring flexibility for teams as they scale their use of autonomous coding assistants.

Functionality is centered around the concurrent execution of coding tasks. Instead of waiting for a single AI agent to complete a series of prompts, Phasr allows users to fan out multiple tasks across independent environments. This parallelization is supported by a robust backend that manages task lifecycles, streaming progress back to the user, and facilitating the integration of diverse AI models. The platform acts as an orchestration layer that adapts prompts and manages context while ensuring that each task remains isolated from others.

Some of the key features are:

• Parallel Execution: Run multiple AI coding agents simultaneously across independent tasks to boost productivity and reduce wait times.
• Git Worktree Isolation: Utilize lightweight, disposable Git worktrees to ensure that each agent operates in its own environment, preventing file collisions and merge conflicts.
• Provider Agnostic: Use the best model for every task, whether it is Claude, Codex, Gemini, or others, without being locked into a single provider.
• Review-First Workflow: Maintain control over the codebase with a purpose-built diff viewer that allows humans to approve, reject, or modify agent-generated code before it reaches the main branch.
• IDE Integration: Seamlessly bridge the gap between agent workspaces and local development by opening tasks directly in preferred editors like VS Code, Cursor, JetBrains, or Zed.
• Task Decomposition: Gain the ability to break down complex, vague features into smaller, manageable units that agents can execute quickly and reliably.

Operationally, Phasr functions as a desktop workspace where users manage their AI agent tasks. When a user launches a task, the platform automatically creates a unique branch and a corresponding Git worktree. The AI agent performs its assigned work within this isolated directory, with progress streamed to the Phasr interface in real time. Upon completion, the changes are surfaced in a clear diff viewer. The user then reviews these changes and decides whether to approve the merge or request modifications. Once a task is finished, the disposable worktree is removed, leaving the repository clean.

Some common use cases include:

• Massive Refactoring: Breaking a large-scale refactor into smaller, independent tasks that can be executed by multiple agents in parallel to save time.
• Automated Test Generation: Running multiple agents to generate unit tests across different modules simultaneously, ensuring high coverage without manual overhead.
• Parallel Bug Fixing: Tasking different agents to investigate and fix unrelated bugs concurrently in isolated worktrees to speed up incident resolution.
• Feature Prototyping: Using different models to explore multiple implementations of a new feature simultaneously, comparing the results to select the most efficient approach.