This Pi extension enhances agent delegation by enabling asynchronous subagent execution with advanced features like chaining, parallel processing, output truncation, artifact management, and session sharing. It empowers users to build sophisticated, multi-agent workflows within the Pi framework.

How It Works

The core mechanism involves defining specialized "agents" via markdown files with YAML frontmatter, specifying models, tools, skills, and execution parameters. These agents can be orchestrated sequentially (/chain) or in parallel (/parallel) using slash commands. The extension supports fine-grained control over agent behavior through frontmatter configurations and runtime overrides, including extension sandboxing and MCP server tool integration.

Quick Start & Requirements

• Install: pi install npm:pi-subagents
• Prerequisites: Requires the pi-web-access extension for certain built-in agents. MCP tools necessitate the pi-mcp-adapter extension.
• Links: No direct links to official quick-start guides or demos were found in the provided README.

Highlighted Details

• Slash Commands: Offers /run, /chain, and /parallel commands with tab-completion and live progress rendering for agent execution.
• Agents Manager TUI: A comprehensive Text User Interface (TUI) accessible via Ctrl+Shift+A for browsing, editing, creating, and launching agents and chains.
• Chain Files: Supports reusable .chain.md files for defining complex, multi-step pipelines with per-step configuration overrides.
• Parallel Execution: Enables fan-out/fan-in patterns for concurrent task execution, with options for concurrency limits and fail-fast behavior.
• Management Actions: Allows LLMs to dynamically discover, inspect, create, update, and delete agent and chain definitions at runtime via the subagent tool.
• Async Support: Provides background execution for single agents and sequential chains, with observability through status files and a TUI widget.
• Recursion Guard: Implements a configurable depth limit to prevent unbounded nested subagent calls.

Maintenance & Community

No specific details regarding maintainers, community channels (e.g., Discord, Slack), or project roadmap were present in the provided README content.

Licensing & Compatibility

The provided README content does not specify the project's license or any compatibility notes for commercial use.

Limitations & Caveats

Parallel execution is strictly synchronous. Asynchronous execution is limited to single agents and sequential chains. Deeply nested subagent calls are restricted by default, though this limit is configurable. Some MCP tool functionalities may require a Pi restart after initial setup.