Shaula

The AI-staffed back office for a therapy practice — website, blog, FAQ, review replies, newsletters, paperwork drafts — run as kanban workflows on your own machine. Nobody hosts your data. This office houses nothing.

Download

Shaula Desktop — the app. Download, open, and the first-run screen walks you through everything (it even installs the free local AI for you). No account, no terminal, no fees, no telemetry.

⬇  Download for Mac

Apple Silicon (M-series) · macOS 12+ · ~120 MB app + a one-time local-AI download on first run · Windows and Intel-Mac builds are on the way. All downloads →

Prefer the terminal?

# 1. clone $ git clone https://github.com/matthewsextonlcsw-sudo/shaula-office.git $ cd shaula-office # 2. one walkthrough — free & offline (Ollama on this machine)… $ bin/shaula-setup --local # …or the cloud brain: Gemini on YOUR Google Cloud project, YOUR billing $ bin/shaula-setup --cloud YOUR_PROJECT_ID
View the source on GitHub Read the full guide

Pick your brain

TierRuns onCostPrivacy posture
LocalOllama, this machine$0Offline. Nothing ever leaves the box.
Cloud brainGemini on your own Google Cloud Vertex projectYour Google billing, pay-per-useYour project, your data terms, your Google BAA if you need one.

What makes it different

An honesty engine, enforced

Every output passes a banned-language gate: no invented statistics, no outcome promises, no fake credentials. The linter refuses to ship them — and re-checks at review.

Human gates, real ones

Anything that would be published, posted, or sent stops on a triage column until a person approves it. A model cannot approve its own review.

17 workflows, day one

Website launch, weekly blog, FAQ page, review replies, content calendar, local presence, welcome emails, social clips, practice paperwork, research desk, and more.

Runs the office, not the therapy

No clinical decisions, no crisis handling — those belong to the clinician, always. Shipped workflows are no-PHI by construction.

Honest requirements

macOS or Linux · Python 3.11–3.13 · git · Ollama (local tier) or the gcloud CLI + your own project (cloud tier) · Node 20+ for the board UI · roughly 2 GB for the harness and 5 GB for the local model. The setup walkthrough checks all of this before it touches anything, and never runs a billable command for you.