An open-source coding and research agent built on a persistent Python REPL, with subagents, a self-refining harness, and support for long-running tasks.
nicobailon/pi-subagents
A Pi extension that lets the Pi coding agent delegate work to focused child agents, run in the foreground or background.
- Stars
- 4,148
- Forks
- 845
- Open issues
- 1
- Last commit
- 48 minutes ago
What is it?
pi-subagents is an extension for Pi, the coding agent. It lets the parent Pi session hand off a task to a focused child Pi session with its own job, then brings the result back into the conversation. Typical uses include code review, codebase scouting, implementation, parallel audits, and saved multi-step workflows.
Foreground children stream their progress inside the conversation. Background children run in a detached runner process, so work continues after control returns to the user and can be checked later. A persistent FleetView below the editor shows active runs, and a `/subagents-fleet` inspector lets you read transcripts, steer a running child, or stop it.
The package ships builtin agents including scout, researcher, worker, reviewer, oracle and delegate, plus council mode for comparing model-based advisors. Users can ask for these in plain language, and custom agents can be defined in their own agent directory.
Key features
- 01Delegates tasks to focused child Pi sessions from a parent session.
- 02Runs child agents in the foreground with streamed progress or in the background.
- 03Includes builtin agents for scouting, research, implementation, review and second opinions.
- 04Shows active runs in a persistent FleetView and an inspector for transcripts and control.
- 05Supports parallel reviewers, saved workflows and council-style multi-model comparison.
Use cases
- Getting a code review of a diff from a separate reviewer agent before merging.
- Scouting an unfamiliar codebase before planning a change.
- Running several reviewers in parallel for correctness, tests and complexity.
- Launching long-running tasks in the background while continuing other work.
Quick start
- 1
Install the extension with Pi
pi install npm:pi-subagents - 2
Ask Pi in plain language to use a subagent, e.g. review a diff with reviewer
Things to know
- Requires Pi 1.2.0 or newer.
- Background children on the standalone Pi build are supported only on Linux x64 per the README.
In collections
Topics
FAQ
- What is pi-subagents?
- A Pi extension that lets the Pi coding agent delegate work to focused child agents, run in the foreground or background. pi-subagents is an extension for Pi, the coding agent. It lets the parent Pi session hand off a task to a focused child Pi session with its own job, then brings the result back into the conversation. Typical uses include code review, codebase scouting, implementation, parallel audits, and saved multi-step workflows.
- Is pi-subagents free to use? What license does it have?
- pi-subagents is open source under the MIT license. Check the license terms before using it in commercial products.
- Is pi-subagents still maintained?
- Actively maintained. The last commit to pi-subagents was on Nov 7, 2026.
- What can I use pi-subagents for?
- Getting a code review of a diff from a separate reviewer agent before merging. Scouting an unfamiliar codebase before planning a change. Running several reviewers in parallel for correctness, tests and complexity. Launching long-running tasks in the background while continuing other work.
- How do I install pi-subagents?
- Install the extension with Pi: pi install npm:pi-subagents Ask Pi in plain language to use a subagent, e.g. review a diff with reviewer
- Where can I get pi-subagents?
- The source code and installation instructions are on GitHub: https://github.com/nicobailon/pi-subagents
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
Enforces your project's AI-agent instruction rules by checking every edit and turn with a decision model and asking the agent to fix violations.
Claude Code plugins that run planned, approved workflows for implementing, designing, reviewing and verifying code changes.
A coding agent built on pi that routes routine decisions, like permission checks and context pruning, to a small judge model so the main model can focus on code.
A local-first, evidence-first AI coding terminal that connects LLMs to real project files, tools, permissions, verification and context.
Self-hosted control center for running coding agents like OpenHands, Claude Code and Codex across local, remote and cloud backends.