Launching today
Joanium
Local AI workspace to build and work with your computer
86 followers
Local AI workspace to build and work with your computer
86 followers
Joanium is a local-first AI desktop app that goes beyond chat. It reads your project files, runs automations on a schedule, operates background agents, and connects natively to GitHub, Gmail, Google Drive, and Calendar, all without sending your data to the cloud. Supports 10+ AI providers including Gemini, Claude, and GPT. Your data stays on your machine. Always.
Joanium
👋 Hi Product Hunt!
I'm Joel, the creator of Joanium.
Joanium is an AI-powered software engineering assistant designed to help developers build, debug, and maintain applications more efficiently. It can assist with coding tasks, project understanding, and day-to-day development workflows.
I built Joanium because I wanted a tool that feels more like a reliable engineering teammate than just another chatbot.
We're launching today and would love your feedback:
What features stand out most?
What would make you switch from your current AI coding tools?
What workflows should we support next?
Thanks for checking out Joanium! I'll be here all day to answer questions and discuss ideas. 🚀
Raycast
Joanium
@chrismessina Great question! Right now, Joanium is focused on building the best local-first experience and the project is free to use.
Longer term, I'm considering a paid tier that would provide managed AI compute and access to premium models for users who don't want to bring their own API keys, while keeping the local-first approach at the core. Still early, and I'm spending more time talking to users before locking in the final model.
@withinjoel Joel, the local-first stance is the right one and the "scheduled background agents" framing is what most desktop AI apps have been avoiding. Two questions that decide whether this is a toy or a daily driver: (1) when a scheduled agent runs while I'm offline and Gmail / Drive have moved on, how do you reconcile the stale view it operated against, and (2) is there a way to inspect what an agent did between runs without me opening every file it touched. Saw a similar local-first shape on a launch a couple of days ago called Second Brain for AI (Cloudflare backed), the failure mode there was conflict resolution across MCP clients writing into the same memory. Joanium will hit a version of that as soon as more than one provider talks to the same project folder.
Joanium
@fabriziowexare Great questions.
For (1), I don't think I have a complete answer yet. Handling stale views and reconciliation across external systems is definitely one of the harder problems for scheduled agents, and it's something I'm still exploring as the product evolves.
For (2), yes — every execution is logged. You can inspect details such as the prompt, response, model/provider used, token usage, timestamps, and other execution metadata without having to manually trace through everything. I've attached a screenshot of the current execution log view.
For (3), today the approach is fairly simple: agents are made aware (through their context/prompt) when other agents are working on the same project. I agree that's not the same thing as full conflict resolution, versioning, or locking, and those are challenges I'll need to tackle as multi-agent workflows become more common.
Really appreciate you bringing these up — they're exactly the kinds of problems that separate a demo from a daily-driver product.
The local-first approach is what caught my attention. Most AI tools talk about privacy, but very few actually keep the data on the machine. Curious how much of the workflow can run fully offline versus requiring API access to external models?
Joanium
@sydneywills114 Thanks Sydney! Joanium is designed to be local-first, so core workflows can run entirely on your machine using local models such as Ollama. In that setup, files, conversations, and AI processing stay on-device and can work offline.
External APIs (OpenAI, Anthropic, Gemini, etc.) are optional and only used when you choose to connect them. Similarly, connectors such as GitHub naturally require network access to communicate with those services, but Joanium itself doesn't require cloud AI services to function. The goal is to give users full control over where their data is processed.
mailX by mailwarm
Is there a way to route different tasks to different models automatically, or do users pick one per chat?
Joanium
@thamibenjelloun We support automatic model routing, currently available in our Preview Channel. Users who opt into preview builds can configure different tasks to be handled by different models automatically. We're actively testing and refining it before rolling it out to all users in the stable release.