🐖 Piggery

Your coding agents are pigs. Piggery is the farm.
Lợn cày tasks

Install · Guide · Changelog

piggery-intro.mp4

Work goes into a pig's trough and waits there until the pig is back. It only counts as eaten when the job is actually done, not when the pig sniffed at it. Every pig also has a pen: its role says who it may talk to and whether it may have piglets (workers). The farm checks the fence itself, so no amount of sweet talk gets a pig through it.

pi, Claude Code, Codex, omp, dsh and opencode pigs all live on the same farm and talk to each other. The farm is one Go binary and a SQLite file, and nothing runs in the cloud.

How it works

flowchart LR
  a["agent on pi"] <-->|mail| farm
  b["agent on Claude Code"] <-->|mail| farm
  c["agent on …"] <-->|mail| farm
  farm(("🐖 piggery<br/>mailbox + gate")) --- shape[["a team layout<br/>lead → peers<br/>peer ↔ peer<br/>…"]]
  shape --> work[/"your tasks and projects,<br/>plowed"/]

Loading

• One mailbox for every agent, whatever its harness. A pi session can mail a Claude Code session, and a worker's answer wakes whoever is waiting for it.
• A gate on every mail and every spawn. It checks them against the team's layout, a small YAML file you pick or write.
• Layouts are files. Eight come built in (see Farm layouts); any other shape is one more file.

Harnesses

Harness	Your sessions	Workers piggery starts	Tested with	Add piggery
pi	yes (extension)	yes (pi --mode rpc)	0.87.1	piggery setup pi
Claude Code	yes (plugin + MCP)	yes (claude -p)	2.1.283	piggery setup claude
Codex	yes (hooks + MCP)	yes (codex app-server)	0.157.1	piggery setup codex
omp	yes (extension)	yes (omp --mode rpc)	18.4.2	piggery setup omp
dsh	yes (dsh web, plugin)	yes (dsh --profile sdk)	0.2.0-rc.1, 0.2.0-rc.2	piggery setup dsh
opencode	yes (plugin)	yes (opencode serve)	1.18.34, 1.19.0	piggery setup opencode

• Workers run on the harness their role names (spawn.harness), else on the harness of the session that founded the team.
• piggery setup alone shows each harness's version and whether piggery is installed in it.

Install

curl -fsSL https://raw.githubusercontent.com/sting8k/piggery/main/install.sh | sh
piggery setup pi       # and/or: claude, codex, omp, dsh, opencode

On Windows (experimental), in PowerShell:

irm https://raw.githubusercontent.com/sting8k/piggery/main/install.ps1 | iex

The script picks the build for your OS and CPU (Linux, macOS or Windows; amd64 or arm64), checks it against the release's checksums.txt, and puts it in ~/.local/bin (%USERPROFILE%\.local\bin on Windows).

• Another directory: PIGGERY_INSTALL_DIR=.... A given release: PIGGERY_VERSION=v0.3.0.
• By hand: download piggery-<os>-<arch> (piggery-windows-<arch>.exe on Windows) from the latest release, chmod +x it, put it on your PATH.
• With Go 1.26+: go install github.com/sting8k/piggery/cmd/piggery@latest.
• Later: piggery update installs a newer release. What changed is in CHANGELOG.md.

Quick start

1. Open pi, Claude Code, Codex, omp, dsh web or opencode in your project.
2. Ask it for a team: "make a lead-peer team to fix the failing tests".
3. Watch the farm: piggery top.

More:

• docs/guide.md: running teams, watching and stepping in, customizing ~/.piggery.
• docs/reference.md: every command, config key, profile key and manifest key.

In Paseo

Using Paseo? PigMonitor shows your teams in the app: who is working, who needs you, each agent's transcript and mail.

paseo plugin install github:Patruxs/PigMonitor

Overview: every team and agent on a timeline, coloured by state	Graph: each team as an org chart, with what each agent is doing

PigMonitor is made by Patruxs, who took piggery further inside Paseo than our own plugin ever did. Thank you! The screenshots are from PigMonitor's README.

Farm layouts

Template	Who does what
lead-peer	A lead owns the plan and judges the results; peers own scopes and speak up with evidence.
slp	You steer a supervisor; each lane has a lead and peers, often in its own git worktree.
council	A chair asks members for independent views on one hard decision.
amp-like	A lead does the work; an oracle (hard reasoning) and a reviewer (diffs) each answer once.
dual-lens	A taskforce: a chair takes one hard question or review, asks two lenses on different models, then makes them answer each other's conflicts.
advisor	A taskforce of one: read-only advice on a hard decision or a stalled approach, with a check that would prove it wrong.
gastown-like	A mayor splits the work; polecats do each task on its own branch; a refinery merges them one at a time.

Each one is drawn, with when to pick it, in manifests/README.md.

Make your own: piggery template new mine --from slp, then edit ~/.piggery/templates/mine/manifest.yaml (see the guide).

Build from source

go build ./cmd/piggery
go test ./...

The tests compare files byte for byte, so every file is checked out with LF line endings (.gitattributes), on Windows too. A Windows clone made before that file covered everything may still hold CRLF copies; check them out again with git rm -rq --cached . && git reset -q --hard.

License

MIT