I ve been using Claude Code as my main terminal driver for a while, but it has a huge blind spot: it operates completely in isolation. Whenever I needed Gemini to do deep repo research, or Cursor to handle UI work simultaneously, Claude had no idea what they were doing. I was stuck playing the middleman, copy-pasting output from one terminal into the other. So I built Neohive. It s an open-source MCP server that gives Claude Code a shared local directory to communicate with your other agents. How it actually works:
* You run `npx neohive init` in your project.
* Claude Code and your other CLIs read/write to a shared `.neohive/` folder on your drive.
* Claude can now autonomously ping Gemini, assign tasks, or ask for code reviews without you copying text. Everything is local. No cloud, no database, just the filesystem acting as a message bus. Repo is here if you want to test it out: https://github.com/fakiho/neohive If anyone else is running Claude Code alongside other local models, I'd love to know how you handle keeping their context synced.
Quick story I'm French, and if you've ever typed in French on macOS or iOS, you know the pain. Autocorrect "fixes" words into things you never meant to say. I once sent "je suis excit " (which has a very different meaning in French than "I'm excited") to a client. That was the last straw.
So my co-founder and I built Exact a tiny AI-powered writing assistant that lives in your menu bar. You press Ctrl+Space in any app and it instantly fixes spelling, grammar and syntax. No copy-paste, no browser extension, no switching tabs. It just works wherever you type emails, Slack, iMessage, even code comments.
We're a team of 3, bootstrapped, and honestly still figuring out a lot. But the core product is solid and we use it ourselves every single day.
Hey everyone I've been building this solo alongside a full-time engineering job for the past few years, and I'd love some real feedback from this community.
It's a 3D avatar coach, powered by an LLM & vison models, that can teach you dance moves currently 200+ dance motions across 5 styles, plus warmups and stretching. You talk to it like you would to any real human coach, and it demonstrates, slows down, and breaks down movements based on what you ask.
It's early some movements aren't as precise as I want them to be yet, and I'm actively working on making the correction sharper. That's exactly why I'm posting here: I'd rather get honest feedback now from people who'll actually try it than polish in isolation.
Excited to share Chilcy, a free AI-powered tool that analyzes your CSV or Excel files in seconds. It provides:
Column types detection
Missing values, outliers, duplicates
A 0 100 data quality score
Whether you re a startup founder, analyst, or data enthusiast, Chilcy helps you quickly understand your data before feeding it into AI models or reports.
Try it here: https://www.chilcy.com/tools/dat... Web app: https://app.chilcy.com/
Shipping this one nights and weekends as a solo dev. It's a golf career management sim, browser-based, free, no ads. You create a golfer, pick an archetype, and play through full seasons against a persistent AI field, training, tournament selection, and career arc all matter over time.
Posted an early version to Reddit a while back and got honestly roasted, 0% retention, confusing UI, bugs everywhere. Fixed what I could, kept iterating, and it's in a much better place now. Sharing here because I want more eyes on it and more honest feedback before I keep building.
I built a free, private, offline focus app that uses Apple Intelligence to understand what you're actually doing.
I was fed up with focus apps that just block a list of websites. I find that blocklists are too "blunt", for example, I could genuinely be using YouTube to research something, but a blocklist sees "youtube.com" and immediately blocks it.
I m launching Ozyplan this Saturday after seeing the same problem with most student planners: they collect every deadline but still leave the student to decide what deserves attention first.
Ozyplan connects to Canvas read-only, imports a calendar feed, or accepts course files. It reads deadlines and requirements, estimates workload and risk, then produces one prioritized daily briefing with the first steps, time estimate, confidence, why that item won, and the alternative it rejected.
The first course analysis and first briefing are free with no card. Pro is $9.99/month after that. I m looking for students willing to test it against a real syllabus and tell me where the first recommendation earns or loses their trust.
