I built this because I feel the vast majority of people do not (and cannot) get the most out of AI at all, or at least without handing their sovereignty over to big tech. The good AI tools are locked behind subscriptions and APIs that watch what you do. And most of what is out there is just a chat box. I wanted something that actually works in my files, reads my email, manages my calendar, and remembers who I am between conversations. So I made it.
It is a native macOS app with a tiny Python backend bundled inside. Clone the repo, open the Xcode project, hit Build. That is it. The backend spawns when the app launches and talks to it over localhost. Nothing leaves your machine.
There is a also a .dmg file in the releases section of the github repo for an easier download process.
- Your data is one SQLite file in
~/Library/Application Support/JoeBro/
. Back it up with cp.
- No telemetry, no account, no phoning home.
- Pick the model. Local Ollama or any OpenAI-compatible endpoint (DeepSeek, Anthropic, Groq, Gemini, OpenRouter, paste a key and go).
- THEMING!! Custom wallpapers behind the glass UI. Solid colour themes too.
The Full Workspace:
- Chat** with live streaming, extended thinking, and real agent mode. Switch models mid-conversation without losing context. Sort into folders. Drag and drop. Deep Research that reads many sources and writes a cited report with images, all on your machine. Documents opened right beside the chat. Real Word .doc and .docx files, edited in place with full formatting.
- Edit any doc/code type you could possibly think, with your agent, directly on your disk in JoeBro.
- Render HTML and SVG in the workspace after editing the code and even open PDFs right there to read with your agent.
Email over IMAP. Read, compose, reply, forward, triage. The agent handles it with real tools, not guessing. Calendar with natural-language quick add (“lunch with Sam Tuesday 1pm”). The agent creates and manages events for you. Brain persistent memory that lives in SQLite. Facts, preferences, project context. The agent remembers across sessions. Tasks scheduled automations. Morning email summaries, weekly reviews. They run as agents on their own. Skills JoeBro learns what you do often and turns it into reusable procedures with confidence scoring. Review, edit, or prune them. AI Check paste text, see how AI-written it reads, suspect sentences flagged.
All info (tasks, skills, memories) are stored locally and editable/deletable directly in JoeBro.
Tools tab - bring your own:
API Tools - point any JSON endpoint at the model. Give it a URL, name, description, optionally an API key and HTTP method. Put {query}
in the URL and the model drops user input right in. Search LinkedIn, Crunchbase, GitHub, weather, HackerNews, anything with a URL works. Toggle on and off anytime.
-MCP Servers - Model Context Protocol over stdio. The app launches the server, discovers its tools and lists them, calls them, then kills the process. Stateless. Hard wall clock timeout so a broken server never hangs a turn. The first launch can take a moment if npx has to download, and errors show in red on the row. No zombie children. Plugins - folders on disk that ship their own tools and agent logic. Two kinds: Foreground (active tools the model can invoke) and Background (guardrails that shape every turn without being called). The bundled macOS Use plugin controls the Mac through osascript and screencapture. No node modules, no Python packages. Background plugins report themselves in the “Plugins used” line at the bottom of the reply.
The backend is standard library Python. Zero pip install commands. One Xcode project, one Build.
The agent calls API tools, memory, tasks, calendar, and plugins in one conversation. Permission modes (bound folder, read-only, full access) control how much of your files it can touch.
Take control of your own AI and get the most of it without submitting to big tech!!
Any and all feedback/questions are appreciated!