All Projects

Live Build · Phase 1

Stewardship Node

A small intentional community on a rural northern Alberta property. Off-grid power, shared food production, a system that remembers.

Here is our community. We proved it works. Please use it to prosper.

The Idea

A handful of people. A piece of land. Greenhouses running through the short Alberta growing season. Permanent tiny house structures. Animals, eventually. Enough land to work.

The part that makes this different from other homesteads is that the community runs on a shared memory. A Meal knows what's in the pantry and what grew well last year. Marlene holds the calendar, the votes, and the task queue — and surfaces the right thing at the right time. Kevin does the morning rounds so no one has to walk the greenhouse in January before the sun comes up. The power system and the network just work, invisibly, the way infrastructure should.

Each of these is designed to stand alone. Each is also designed to talk to the others. Together they make the community smarter over time — without requiring anyone to hold the whole picture in their head.

None of this replaces the human part. It makes room for more of it.

How It Learns

The Feedback Loop

Last October, the winter squash yield came in higher than expected — more than the community could eat fresh. Here is what happened next.

01

A Meal

tracks overproduction

Harvest logs showed winter squash yields 40% above consumption baseline. A Meal flagged it automatically.

02

Marlene

suggests recipes

Three options surfaced, ranked by what the pantry could support. The experimental ferment was option three.

03

The Community

votes

All-time high vote of no on the ferment. The preference is now in the record — not in anyone's memory.

04

A Meal

adjusts next spring

Planting scale for that variety comes down. The community's collective preference becomes next year's growing plan.

The community's collective preference becomes next year's growing plan. The system remembered so no one had to.

The System

The Characters

A Meal

Knows what's in the pantry. Remembers what grew well last season. Noticed the overproduction before you did.

A Meal is the community's relationship with food made visible. Every greenhouse sensor, every harvest log, every jar counted on the shelf — it holds all of it, and it connects the dots across time.

When October's squash yield comes in higher than expected, A Meal flags it. When the community votes no on a recipe at an all-time-high rate, A Meal remembers. When spring planting season arrives, that memory becomes a recommendation: grow less of that variety this year. The feedback loop closes quietly, automatically, without requiring anyone to hold it in their head.

You don't need to be an engineer to use it. You need to be someone who cares what's in the pantry.

Under the hood
  • AWS IoT Core + MQTT for greenhouse sensor telemetry
  • DynamoDB for harvest records, pantry inventory, vote history
  • Lambda processing functions + API Gateway
  • SQLite edge buffer with async sync — works through network outages
  • Marlene integration for anomaly escalation and planting recommendations

Marlene

Holds the threads. Surfaces the right thing at the right time. Coordinates — does not decide.

Marlene is the connective tissue of the community's operational life. She watches A Meal's telemetry and Kevin's task queue. She knows the maintenance schedule, the harvest calendar, and the list of things someone said they'd get to that haven't happened yet.

When the greenhouse temperature dropped unexpectedly at 2am last January, Marlene flagged it and put a task in Kevin's queue for the morning round. She didn't fix it — that wasn't her call. She made sure the right people knew, with enough context to act.

She suggests recipes based on what's actually in the pantry. She tracks the votes. She adjusts the planting recommendations when the community's preferences shift. She is never the decision-maker. She is the reason you have enough information to make a good one.

Under the hood
  • Claude API with OPA-enforced scope — defined domains, hard limits on autonomous action
  • A Meal telemetry + Kevin task interface as primary inputs
  • Plain language interface designed for non-engineers
  • Covenant-style policy layer: Marlene cannot act outside her domain without explicit confirmation
  • Audit log of every recommendation and its outcome — the vote history is part of her memory

Kevin

Walks the property so you don't have to. Every morning. In January. Before the sun comes up.

Kevin does the rounds. He checks the greenhouse sensors in person — not just whether they're reporting, but whether something looks wrong that the data hasn't caught yet. He reports back. He does it again tomorrow.

He is not a general-purpose robot. He is purpose-built for this property, this climate, these tasks. He handles the recurring physical work that piles up as the community grows — the work that matters enormously but doesn't require a human to do it.

When Kevin surfaces an anomaly during a morning round, a community member makes the decision. Kevin does the walking.

Under the hood
  • ROS 2 Humble on Jetson Orin NX 16GB
  • Unitree locomotion + Dynamixel XM540 arm actuation
  • SROS2 security baseline — identity-first before any application logic
  • Draws from the community 48V LiFePO4 bus — same power system, same BMS monitoring
  • X.509 certificate identity provisioned via the same Terraform infrastructure as every other node
  • Simulation-first development: Isaac Sim validates behaviour before hardware build

The Backbone

The lights stay on and the network stays up. You don't have to think about it.

The power system runs on a 48V LiFePO4 battery bank with solar as primary generation. It is sized for Alberta winters — overcast stretches, sub-zero temperatures, short peak sun windows — because those are the conditions that actually matter. Every load is metered. The community knows what's drawing from the bus and when.

The network is a WireGuard mesh across every structure on the property. Whether the connection between two units is a buried cable, a point-to-point wireless bridge, or an LTE failover link, the address space is flat and encrypted. A sensor in the far greenhouse looks identical to a sensor in the room next door.

The orchestration cluster keeps running through network partitions. Services don't stop because the internet went down. When connectivity returns, state syncs. The community doesn't experience this — it just works.

Under the hood
  • 48V LiFePO4 battery bank — 3,000–5,000 rated cycles, cold-tolerant chemistry
  • Solar primary, grid/generator backup sized for Alberta overcast seasons
  • All loads instrumented — power consumption is telemetry, not a guess
  • WireGuard mesh across all property nodes — flat encrypted address space
  • Ubiquiti point-to-point for structure-to-structure physical links
  • VLAN segmentation: IoT devices, residential compute, infrastructure, and management are separate broadcast domains
  • Nomad SCDC cluster on low-power dedicated hardware — holds state through partitions, reconciles on reconnect

The Build

How We Get There

Four phases. Each one proves itself before the next begins. Nothing is claimed before it exists.

Now

Phase 1 — The Workbench

Before there is a community, there is a workstation. A place to build Kevin's software before Kevin exists. A place to simulate the greenhouse network before there is a greenhouse. Phase 1 is not exciting to describe — it is the enabling condition for everything else. It is happening now.

↓ Build checklist

Workstation

In Progress
  • Components selected
  • Build complete
  • Isaac Sim validated — Kevin simulation environment confirmed
  • Development environment configured — ROS 2, Terraform, LocalStack, cluster nodes

A Meal — Core Build

In Progress
  • Architecture finalized
  • Lambda + DynamoDB stack deployed
  • IoT Core device registry live
  • Offline edge agent complete (SQLite buffer + async sync)
  • Marlene anomaly integration

Kevin — Software Foundation

In Progress
  • Architecture designed
  • Phase A: Docker L4T, colcon workspace, systemd
  • Phase B: URDF, interface definitions
  • Phase C: SROS2 security baseline
  • Phase D: core nodes, CI/CD, simulation validation
Planned

Phase 2 — The First Unit

Sam's mobile unit is the proof of concept. A self-contained, off-grid tiny house for a graduate student in Bio-Chemical Engineering — mobile enough to move, connected when co-located, fully independent when not. We find out here whether the design actually works before committing to permanent structures.

↓ Build checklist

Mobile Unit

Planned
  • Floor plan and structural design complete
  • 48V power system sized and spec'd
  • Build complete

Community Network — Node 1

Planned
  • WireGuard peer deployed
  • Nomad client node joined to cluster
  • IoT sensor baseline live (power monitoring, environmental)

A Meal — First Field Deployment

Planned
  • Edge agent deployed to unit hardware
  • IoT Core sync validated from mobile unit on LTE
  • First harvest records logged
Planned

Phase 3 — The Property

Land purchase. First permanent residential units. The greenhouse. The full power system. This is the phase where the community stops being a plan and becomes a place. Everything built in Phase 1 and 2 gets deployed into the real world.

↓ Build checklist

Land and Site

Planned
  • Site identified and assessed
  • Land purchased

Power and Network

Planned
  • System sized to Phase 3 load budget
  • Array and battery bank installed
  • All structures on WireGuard mesh
  • All loads instrumented

A Meal + Marlene — Full Deployment

Planned
  • Greenhouse sensor network deployed
  • Full telemetry pipeline live
  • Marlene coordination layer active
  • OPA access policy defined and enforced
  • First community vote recorded in the system
Planned

Phase 4 — Kevin Comes Home

Kevin moves from simulation to the property. The fleet grows. The community has been running its own infrastructure long enough to know what works, what doesn't, and what to hand to the next people who want to build one of these.

↓ Build checklist

Kevin — Hardware Build and Field Deployment

Planned
  • Jetson bring-up and sensor integration complete
  • Outdoor navigation validated on property terrain
  • Agricultural task capability proven
  • Field deployment — morning rounds begin

Fleet and Handoff

Planned
  • Second Kevin unit hardware build
  • Fleet coordination through Marlene validated
  • All systems documented for the next community

For the Next One

Built to be handed off.

Every component in this system is infrastructure-as-code, documented, and designed for the person who inherits it — not the person who built it. The goal is not a community that requires its founder to keep running. The goal is a community that runs.

When Phase 4 is done, the runbooks, the IaC, the architecture decisions, and the reasons behind them will all be here. If you want to build your own Stewardship Node, you will not start from scratch.

Part of the Same System

Kevin — Steward Robotics Platform

The morning rounds. The physical action layer.

View →

Nomad Edge

The network topology the community backbone runs on.

View →

Watershed

The offline-first IoT pattern A Meal's edge agents follow.

View →