01
Apps
Opinionated by default
Web, mobile and desktop apps with strong points of view. We sweat the details until the interface feels obvious — then sweat them again.
App & agent studio · est. 2026
Stubbornly good software.
SoftwareDonkey builds apps and private AI agent platforms — and teaches small teams to build with AI themselves. Sturdy engineering, and a flat refusal to leave anything half-done.
The job
We do the donkey work.
SoftwareDonkey is an independent app and agent studio. What we build is carried end to end — imagined, designed, built, and hauled over the finish line — and what we know, we teach.
02
Agents
Sturdy beats shiny
Private agent platforms that take on real work — running on hardware you control, sharing what they know, and built to be checked. Boring‑on‑purpose engineering, not this week’s trend.
03
Training
Carried all the way
Hands‑on training in building with AI — how to scope the work, direct the tools, check what comes back, and ship it. A small team leaves able to carry the load itself.
The herd
Meet the donkeys.
SoftwareDonkey runs on its own software. Six agent personas, each with exactly one job and no opinions about anyone else’s — administration, facilities, maintenance, health, finance and security. They are how a small studio carries a large load.
Atlas
Administration 01Atlas carries the paperwork — the inbox, the calendar, the renewals, everything with a date attached. It works the front door so that nothing important quietly rots in a folder.
Flint
Facilities 02Flint manages the buildings — heat, power, water, occupancy, the dull telemetry nobody reads until the day it matters. It writes each place a daily journal so the building has a memory.
Jack
Maintenance 03Jack does the maintenance on client sites: plugins, themes, core, backups, patches, reboot — in that order, every time. It only calls an update done when the version actually moved.
Ridge
The trail 04Ridge is the coach — walking, hiking, yoga, the bike. Its own iPhone app brings Apple Health data home to a private, on‑premises server, so it reads the real training history instead of guessing at it. Cheerfully unimpressed by excuses.
Fin
The ledger 05Fin keeps the books — the overview, the strategy, and the bookkeeping nobody volunteers for. It shows its work and asks first, because that is the only way to trust a number.
Ward
Security 06Guard donkeys are a real thing — farmers put one in with the sheep because it brays at coyotes. Ward reviews security events with analysis accelerated on local GPUs, speaks up when something warrants it, and stays quiet the rest of the night.
ONE DONKEY, ONE JOB. THAT’S THE WHOLE ORG CHART.
The model
A bench, not a bot.
One clever assistant that knows a little about everything is a party trick. A standing bench of specialists — each one resident in its own domain, on watch, with a memory of what happened last time — is an operation.
The right expert, not the nearest one
Different work wants different judgment. Each donkey lives in one domain full-time and keeps its own memory of it — and the whole herd reads from one curated knowledge base, kept in sync in real time. What one donkey learns, the next one already knows.
Somebody is already watching
Because a donkey is resident in its domain rather than summoned into it, it notices before you think to ask. Drift, a stalled job, a number that moved the wrong way, a door at the wrong hour — the first you hear of it is the thing itself, not a report you had to go run.
Ask it, or just look
Talk to a donkey in plain language and get a straight answer. Or skip the conversation entirely — the work gets published as journals, ledgers, summaries and dashboards, so the data is already sitting there when you go looking. Not every question should cost a chat.
It works while you don’t
Hand a job off and walk away. It runs on its own schedule, at night, across the weekend, in parallel with five other jobs, and comes back when it is finished or when it genuinely needs a person. You stop being the bottleneck and you never become the babysitter.
The long, dull, many‑system jobs
The work that flattens a week is rarely hard — it is just long, and spread across everything. A donkey plugs into the tools a business already runs — mail, documents, calendars, project boards, ticket queues, the books — and hauls one job across all of them in a single pass.
Checkable by default
A donkey shows its work. Changes get previewed before they are applied, bulk work gets a dry run first, and private data stays on hardware you control. The goal was never an agent you trust blindly — it is one you can check.
From the field
What that looks like in practice.
Four of the jobs the herd runs today. Not demos — the actual standing work.
A maintenance window nobody sat through
Plugins, then themes, then core, then backups, then OS patches, then the reboot — across a fleet of client sites, in that order, each step gated on the one before it. The donkey does not mark an update done because a command exited zero; it marks it done when the version actually moved. What arrives at the end is the exception list, not the transcript.
A building that keeps its own notes
Temperature by zone, power draw, how long the lights ran, when the place was occupied — hundreds of readings collapsed every day into a short journal a person can actually read. A year later, “was last February like this?” has a real answer, because something was writing it down the whole time while nobody was reading.
Books that show their work
Categorize, reconcile, chase down the receipt that goes with the charge, and keep the personal side and the business side from bleeding into each other. Every proposed change is previewed before it is applied and every bulk pass is dry‑run first — because the point of a financial agent is not that you trust it, it is that you can check it.
Health data that lives at home
Health metrics are about as private as data gets, so Ridge got its own iPhone app. It syncs Apple Health data over a private network into an on‑premises MCP server — no third‑party health cloud holding the history — and the coach reasons over what actually happened: steps, workouts, recovery. For a fully private setup, the coach can run on local open‑source models, small enough to run on a single Mac, so nothing leaves your hardware. An agent platform, end to end: the app, the API, the server, and the donkey.
FULL CASE STUDIES ARE BEING WRITTEN UP. THE SHORT VERSION: THE WORK GETS DONE WHILE YOU ARE DOING SOMETHING ELSE.
Get in touch
Talk to a human.
The donkeys do the hauling. A person still answers the mail — so if you have work that needs carrying, or a team that wants to learn to carry it, say so.
Custom software, cloud or infrastructure work? The studio is run by Jacob Nollette, a cloud and software engineer.