One estate, many rows
A real vineyard is divided into parcelles — distinct plots, each worked on its own rhythm, all part of one estate. Pinard borrows the word for a workstream: a named grouping of related issues, workers, and a target branch, tended by its own conductor.
A parcelle is defined by a file at parcelles/<name>/parcelle.yaml:
name: data-pipeline
project: genomics
status: active
description: "Nightly genome build + QC"
target_branch: cuvee/data-pipeline # cuvée routing for this workstream
issues: [42, 43, 44] # issues that belong here
spec: openspec/specs/data-pipeline/parcelle.md
The daemon uses the issues list (and parcelle:<name> labels) to route work, and target_branch to batch each parcelle’s MRs through its own cuvée.
The harvest crew — three tiers
Pinard isn’t one agent; it’s a crew, each with a role:
Régisseur — the estate manager: the general lane, overall coordination
└── Maître — a cellar master per parcelle: conducts one workstream
└── Vendangeur — the harvester: a worker doing one task
- Régisseur — the top-level conductor for the whole vignoble.
- Maître — a per-parcelle conductor window; open or focus one on demand.
- Vendangeur — an individual worker (agent) on a single task.
attach_parcelle("data-pipeline") # spawn/focus this parcelle's maître
Why it scales
Isolating work into parcelles means many efforts run in parallel without stepping on each other: each has its own cuvée branch, its own conductor, its own set of issues, and its own view in the control room and dashboard. Add a parcelle, and the estate grows another row — no reorganization required.
Terroir at every level
From the estate manager down to the harvester, everyone knows the land they work. Parcelles give that structure to your automation: distinct plots, one vineyard, a single harvest.