Most automation tools run your bots on someone else's servers — a foreign IP, a foreign session, a fast track to getting your accounts flagged. Automation Studio runs everything on your own computer, from your own browser session and IP. That makes it the safe way to automate account-bound sites (LinkedIn, dashboards, internal tools) — the work looks like you, because it runs as you.

One painless install gives you three things that compose into something much bigger than the sum of its parts:

🌐 a human-grade browser  ·  🗄️ a built-in data layer  ·  🤖 AI agents that operate both

Out of the box, with the LinkedIn growth suite as the flagship — three workflows that chain into one pipeline: search → connect → message.

Build your own just as easily: drop a Python module in automations/, register it, and it shows up in the app — list-consuming and chainable by default.

Agents are the headline act. They run on your own Claude Code or Codex CLI — your existing subscription, no API keys, nothing extra to pay. They act through the same tools you do (workflows, the data layer, and the live browser), so they can genuinely operate the studio, not just chat about it.

Four specialists ship built-in:

• 🛠️ Studio Agent — the complete operator. Builds & runs workflows, shapes the data layer, drives the browser, schedules work. Point it at anything.
• 💼 LinkedIn Specialist — deep LinkedIn mastery: the search→connect→message pipeline plus live navigation, fluent in all of LinkedIn's quirks.
• 📈 Client Acquisition Lead — a B2B business-development partner. It sharpens your ideal customer, finds prospects across LinkedIn, Google/Maps, websites, directories and niche sources, enriches them with verified contact paths, runs approved multi-channel outreach and follow-up, and advises on sales strategy.
• 📣 Social Growth Lead — a social media growth operator. It plans organic or paid distribution across LinkedIn, Reddit, X, Instagram, TikTok, Facebook and niche communities, publishes approved channel-native content, monitors real traffic and engagement metrics, and keeps iterating over hours or days.

Roll your own in seconds: give an agent a name, a personality/skill prompt, and a toolset — it's ready to launch.

Grab the installer for your OS from the latest release:

_{First launch: builds are ad-hoc / self-signed, so on macOS right-click → Open once, and on Windows click More info → Run anyway.}

┌──────────────────────────── Automation Studio.app ─────────────────────────────┐
│                                                                                 │
│   Electron + React UI  ──HTTP──▶  Python backend (frozen, fully autonomous)     │
│                                     ├── orchestrator/  FastAPI · runs · agents   │
│                                     ├── humanbrowser/  the stealth Chrome engine │
│                                     ├── automations/   the workflows             │
│                                     └── studio.sqlite  the data layer            │
│                                                                                 │
│   AI agents ◀──MCP tools──▶ workflows · data · live browser                     │
└─────────────────────────────────────────────────────────────────────────────────┘

• Local-first by design. The UI is a thin shell over a self-contained Python engine. Everything — browser profiles, datasets, runs — lives in your OS user-data folder, never in the cloud and never in the app bundle.
• Bring your own model. Agents spawn your local Claude/Codex CLI and inherit your login. The only thing that leaves your machine is the model traffic you already pay for.
• Autonomous backend. The engine runs with or without the UI — script it, or let an agent drive it headless.

# backend engine (its own virtualenv)
cd backend && python3.13 -m venv .venv && .venv/bin/pip install -e .
cd .. && npm run frontend:install        # app shell deps

# run in dev (opens Electron, spawns the Python backend, hot-reloads the UI)
npm run dev

# package a native desktop app for this OS (run from the repo root)
( cd backend && .venv/bin/pyinstaller --noconfirm automation-backend.spec )   # freeze the backend
PLAYWRIGHT_BROWSERS_PATH=frontend/resources/chromium backend/.venv/bin/patchright install chromium  # bundle Chromium
npm run build:mac      # or build:win / build:linux, run on that OS

Tagging a vX.Y.Z release on main triggers CI to build & publish installers for all four targets automatically.

Add a workflow by dropping a module in backend/automations/ — drive the browser via humanbrowser.session, emit progress through automations/_events.py, write a CSV — and register it in orchestrator/registry.py. Make it list-consuming with an input contract and it instantly chains with everything else. The UI and the agents discover it automatically.

• Runs on your machine, your IP, your browser session — nothing is proxied through a third party.
• Browser process trees are owned and reaped as a whole (via psutil), so automations can never orphan or accumulate.
• You're automating real accounts: Automation Studio paces like a human and respects platform limits — use it responsibly.