Pi Subagents: Run Multiple AI Agents

Run multiple AI agents in one Pi session: scout does recon, researcher hits the web, worker edits code. Depth-capped. Bring your own Pi model key.

NewNew1.8 MB snapshotStarter VM

What's inside

Harness

Pi

Plan

Starter

vCPU

1

Memory

2 GiB

Snapshot

1.8 MB

How it works · ~3 minutes

  1. 01 · Fork

    New isolated microVM on your subdomain — creator agent state included.

  2. 02 · Your keys

    Log into Pi with your own model credentials (BYOK).

  3. 03 · Ask it to work

    Open the terminal and give the agent a real job. You keep what ships.

Example first ask

Pull my open PRs, summarize risk, and draft a review checklist.

30-day money-back on your first purchase · no free trial · keys never leave the VM

About this template

Three agents. One Pi session. No second orchestrator.

One Pi session is serial. Recon, then web research, then the edit — each waiting on the one before it while you watch.

This fork pre-wires three subagents that run as parallel Pi processes: scout for codebase recon, researcher for the web, worker for the code change. You keep the conversation you were already in.

No n8n. No CI runner. No second harness.

Run multiple AI agents from one prompt

Set a model key, start Pi, and hand it a task that obviously wants three heads:

Scout the auth package, research the library's current recommended pattern, then have worker implement the fix.

Scout, researcher and worker fan out from there. Their work renders inline, in the same terminal, while it happens — you are not tailing a log in a second window.

What lands in your fork

AgentJob
scoutFast codebase recon — read, grep, find
researcherWeb research
workerCode changes
  • A single subagent tool surface — one thing to invoke, not three
  • Depth caps, so a subagent spawning a subagent can't recurse away with your model budget
  • Live inline rendering of child work, inside the conversation

What you must supply: your own Pi model key

Nothing is baked into this template.

export ANTHROPIC_API_KEY=...   # or OPENAI_API_KEY / GEMINI_API_KEY
pi

That is the whole setup. The rig is already installed.

Your first five minutes

Fork it. Export a key. Start Pi.

Then give it one real task from a repo you know well, phrased as recon → research → edit, and read what each agent came back with. If scout names the right files, you have your answer.

Honest limits of this multi-agent rig

This is the Pi harness. If you need Claude Code, Codex or OpenClaw specifically, this is the wrong fork.

Depth caps are the point, so you will not get an unbounded swarm out of it. It also wants a repo on disk — pure chat with nothing to scout gets you very little of the value. And three agents burn three agents' worth of tokens, on your key, on your bill.

Fork it, fan one real task across scout → researcher → worker, and see whether serial ever feels acceptable again. Your first purchase carries a 30-day money-back guarantee.

Inside this fork

Forking copies this template into a brand-new, fully isolated microVM on your own subdomain. Here's exactly what lands in it.

  • Pi agent

    The upstream harness, pre-installed — same version the creator ran.

  • Starter VM

    1 vCPU · 2 GiB RAM · 10 GiB disk.

  • Creator's /persist data

    The captured persist volume is copied byte-for-byte into your fork.

  • BYOK — your keys, your VM

    Add your model API keys after forking; they live only inside your microVM.

Your turn

Your own Pi Subagents: Run Multiple AI Agents, live in about 3 minutes.

Forking copies this Piagent into a brand-new, fully isolated microVM on your own subdomain — the creator's /persist state and all. Add your own keys after it boots; they never leave the box. Don't love it? Your first jurniti purchase comes with 30 days to get every cent back.

New paid VM · BYOK · 30-day money-back on your first purchase · ~3 min to provision

Starter · fork

$25/ mo