A local-first desktop app that lets you compose AI agents from your own Codex / Claude CLIs into scheduled workflows — all your data stays on your machine.
Connects to your own accounts —
Reach your team from your phone —
Pick your platform from the latest release:
| Platform | File |
|---|---|
| macOS (Apple Silicon) | Ordinus-<version>.dmg |
| Windows (x64) | ordinus-<version>-setup.exe |
Builds are unsigned for now, so your OS will warn you before launching. The app is fully local-first — nothing is uploaded anywhere.
The DMG is unsigned, so macOS Gatekeeper will refuse to open it with a misleading "Ordinus is damaged and can't be opened" message. The app is not damaged — Gatekeeper just blocks unsigned apps that carry the quarantine attribute set by your browser.
-
Open the
.dmgand dragOrdinus.apponto theApplicationsshortcut (don't double-click the icon inside the DMG window — that runs it from a read-only volume). -
In Terminal, strip the quarantine attribute, then launch:
xattr -dr com.apple.quarantine /Applications/Ordinus.app open /Applications/Ordinus.app
You only do this once. After the first launch, the app opens normally.
SmartScreen will show "Windows protected your PC". Click More info → Run anyway.
The installer puts Ordinus under Program Files and creates a Start Menu / Desktop shortcut. Subsequent launches open without prompts.
Agents. Compose roles (article writer, code reviewer, security engineer, …) backed by your local AI CLIs. Each agent has a profile, capabilities, and its own conversation space.
Workflow Designer. Visually wire agents and tasks into a DAG. Compiles down to a runnable plan; reuse the same engine across manual runs and schedules.
Workboard. Watch tasks execute, inspect provider output, and steer runs in flight.
Give your agents the tools you already use — each connection runs under your own account, with credentials stored encrypted on your machine.
Your accounts, run locally
- Google — read Gmail, Calendar, and Drive (sending email or creating events is opt-in, off by default)
- WhatsApp — read incoming messages and reply, via device pairing
- X (Twitter) — read your timeline and mentions; posting is opt-in
- LinkedIn — read profiles, companies, and posts
Hosted services (OAuth)
- Linear, Notion, Atlassian (Jira / Confluence), Canva, Datadog
Reach your team from anywhere
- Telegram — pair your own bot to message your agents from your phone and get the work sent back, no laptop needed (Settings → Remote Access)
Nothing about your data leaves your machine — Ordinus never collects, uploads, or shares it. Tools that act on the outside world (send a message, post, create an event) ship disabled; you turn each one on yourself from Settings → Connections.
- Your prompts, conversations, and workflow definitions live in a local SQLite database — no cloud account, no per-seat pricing.
- Providers run as local CLI processes (Codex, Claude). Ordinus orchestrates them; it does not proxy your tokens through a third-party service.
- No telemetry. No phone-home.
Requires Node.js 22.13.0+ and the platform toolchains electron-builder needs.
git clone https://github.com/muratgur/ordinus.git
cd ordinus/app
npm ci
npm run dev # run in dev mode
npm run build:mac # produce a .dmg under app/dist
npm run build:win # produce a .exe installer under app/distFor Windows builds on a non-signing machine, use npm run build:win:local.
Real projects built end-to-end with Ordinus — original prompt, agent crew, workflows, and what came out.
Shape Survivor — a complete Godot Brotato-like with 6 characters, 7 weapons, 8 waves, and a boss, built in ~12 hours by a 6-agent crew across 142 tasks. Zero human-written code.
Ordinus is pre-1.0. APIs, schemas, and the UI may change between releases. The 0.x line is a public showcase — feedback welcome, but breaking changes will happen without ceremony.
This project is developed AI-assisted using Claude Code. The .claude/skills/ directory contains the project-specific skills (architecture guides, IPC contracts, secure-boundary rules, etc.) that shape how the AI collaborates on the codebase. If you're curious about agent-assisted development workflows, those files are worth a read.
See CONTRIBUTING.md. Small, well-scoped changes are easiest to land.
See SECURITY.md for private vulnerability reporting.
MIT © 2026 Murat Gür