The AI desktop assistant that actually lives on your machine.
Multi-model chat · Scheduled automations · Background agents · MCP · Real integrations
🌐 Website · 📖 Docs · 🛍️ Marketplace · ⬇️ Download · 📋 Changelog · 💬 Discussions · 🐛 Feedback
Joanium is a local-first AI desktop agent that actually does stuff — not just talks about doing stuff.
You know how every AI tool is just a really fancy chat box? You type. It replies. You lose context by tomorrow and you're back to copy-pasting your entire codebase into a prompt like it's 2023. Every. Single. Morning.
Yeah. We got tired of that too.
Joanium lives on your machine. It knows your files. It runs jobs while you sleep. And your data? It never leaves your device. Not even a little bit.
Every morning at 8am, Joanium pulls my GitHub issues, summarises them with AI, and drops a priority list into my Slack. I set it up once. It just runs. Forever.
That's not a chatbot. That's an actual agent.
🧠 Multi-model, zero lock-in Switch between Gemini, Claude, GPT, Grok, Mistral, or go fully offline with Ollama / LM Studio. 10+ providers. Use whatever model won the benchmark this week — we genuinely don't care.
🤖 Background Agents Reusable prompts that run on a schedule. Daily code reviews. PR summaries. Changelog monitoring. Happening in the background, without you touching a thing. Set it once, profit forever.
💬 Channels Reply to Telegram, WhatsApp, Discord, Slack, Zulip, and Mattermost — all through AI. Your assistant meets you where you already are.
🔌 Real integrations (not wrappers) GitHub, Gmail, Google Drive, Calendar, Sheets, Docs. These aren't fake pass-throughs — they're first-class tools the AI can invoke naturally, mid-conversation.
🧩 MCP Support Plug in any MCP-compatible server and Joanium instantly gets new superpowers. The ecosystem is exploding in 2026 and Joanium keeps up.
🎭 Personas & Skills Drop in a "senior code reviewer" and your AI thinks differently, talks differently, reviews differently. Skills are markdown files you write once that teach your assistant exactly how you work.
🧠 Long-term Memory Remembers things across conversations. Not in a creepy way. In a "finally, an AI that doesn't have goldfish memory" way.
🗂️ Project Workspaces Scope your AI's context to a specific codebase or folder. No more explaining your project structure from scratch every session.
🔒 Fully offline, fully private
Zero internet. Zero API keys. Zero cloud. Zero excuses. Everything stays in your Data/ folder, on your machine, under your control.
Look, it's 2026. There are a lot of AI agents. Here's the honest breakdown:
| Joanium | Codex | Claude Code | OpenHuman | OpenClaw | Hermes Agent | |
|---|---|---|---|---|---|---|
| Runs on your machine | ✅ | ❌ cloud | ❌ CLI/cloud | ✅ | ❌ cloud-only | |
| Your data stays private | ✅ always | ❌ | ❌ | ✅ | ❌ | |
| Works fully offline | ✅ Ollama/LM Studio | ❌ | ❌ | ❌ | ❌ | |
| Multi-model (10+ providers) | ✅ | ❌ GPT-only | ❌ Claude-only | |||
| Desktop GUI app | ✅ | ❌ web/IDE | ❌ terminal | ❌ chat only | ❌ | |
| Background scheduled agents | ✅ | ❌ | ❌ | ❌ | ❌ | ✅ cloud-only |
| Personas & Skills system | ✅ | ❌ | ❌ | ❌ | ||
| MCP support | ✅ | ✅ | ❌ | ❌ | ❌ | |
| Real integrations (Drive, Gmail…) | ✅ | ❌ | ❌ | ❌ | ||
| Messaging channels (Slack, Discord…) | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ |
| Long-term memory | ✅ | ❌ | ❌ | ❌ | ❌ | ✅ cloud |
| Open source | ✅ Apache 2.0 | ❌ | ❌ | ✅ | ✅ MIT | |
| Free (bring your own key) | ✅ | ❌ | ❌ | ✅ | ✅ |
tl;dr — Codex and Claude Code are coding agents for the terminal. Hermes lives in the cloud. OpenClaw handles chat. Joanium is the only one that's a full desktop app, multi-model, offline-capable, open source, integrates with your real work tools, and talks to your messaging apps. Different category.
Download the app → One click. You're in. Go.
The Skills and Personas in this repo come pre-installed with Joanium — they're your out-of-the-box starting point.
But the real fun is at joanium.com/marketplace — a separate community repo where you can browse, create, and publish your own. Drop in a "startup growth hacker", a "brutally honest code reviewer", a "3am rubber duck debugger" — or build one that's uniquely yours. No code required, just show up and create.
PRs are genuinely welcome — not in the "we value all contributions" corporate way, in the actual "please help us make this better" way.
Read CONTRIBUTING.md to get started. Check SECURITY.md for responsible disclosure. Follow the Code of Conduct.
Hit ⭐ star. It takes less than a second and means everything for an open source project.
Share it. Tweet it. Tell your developer friends who are still copy-pasting context into ChatGPT every morning. They need this.