tintinweb/pi-subagents
Claude Code like Sub-Agents & Workflow Orchestration for Pi — parallel execution, live widget, fleet view, custom agent types, mid-run steering, claude compatible dynamic workflows and more ...
What it solves
pi-subagents is an extension for the pi platform that enables the creation and orchestration of autonomous sub-agents. It allows users to spawn specialized agents with their own tools, system prompts, and models to handle complex tasks in the background or in parallel, reducing the cognitive load on a single main agent.
How it works
The extension introduces an Agent tool that the main model can use to spawn sub-agents. These agents can run in isolated sessions—either in the background (default) or blocking the main session. Users can define custom agent types via YAML frontmatter in Markdown files, specifying system prompts, model selection, and tool restrictions.
Key orchestration mechanisms include:
- Scripted Workflows: A
SubagentWorkflowtool that executes deterministic JavaScript scripts to manage agents via patterns likeparallel(),pipeline(), andphase(). - Agent Mentions: Using
@handlesyntax at the prompt to communicate with, resume, or start agents without consuming turns in the main conversation. - FleetView: A navigable UI list below the editor to monitor and jump into live agent conversations.
- Git Isolation: The ability to run agents in isolated repository copies, with changes auto-committed to branches upon completion.
Who it’s for
Developers using the pi platform who need to manage multiple specialized AI agents for complex software engineering tasks, specifically those who want a Claude Code-style experience of autonomous sub-agent orchestration.
Highlights
- Parallel Execution: Spawn multiple background agents with configurable concurrency limits.
- Custom Agent Types: Define specialized agents with custom prompts and toolsets via project or global config files.
- Deterministic Workflows: Run JavaScript-based orchestration scripts compatible with Claude Code's
Workflowtool. - Live Monitoring: A persistent widget and FleetView for real-time tracking of token counts, tool activity, and agent status.
- Mid-run Steering: Inject messages into running agents to redirect their work without restarting them.
- Session Resume: Pick up where an agent left off by preserving full conversation context.
- Scheduling: Support for cron, interval, and one-shot scheduling of agent tasks.
Related
- Project
- Project
- Project
- Project
- Project