ZYRAXON AI - A desktop AI agent that plans, acts and verifies
by•
Source-available desktop AI agent with 888+ tools across 63 categories, 11 agent modes, screen vision and eternal memory.
Replies
Best
Hunter
📌
Hi Product Hunt! 👋 I'm Sayidi, a student from Bangladesh, and I built ZYRAXON AI on my own, with one laptop and no team.
Why I built it
Most AI tools answer questions. I wanted one that actually does the work: you give it a goal, and it plans the steps, picks the right tools, runs them, reads the results, fixes its own mistakes, and checks that the job is really done.
What's inside
- 888+ built-in tools across 63 categories
- A catalog of 8,202 MCP servers, each labeled with how far it was tested. Every server connects the way it needs to: nothing to do, a browser sign-in, an API key, or your own URL
- 11 agent modes (plan, build, explore, auto and more)
- Screen vision, so the agent can see what's on your screen
- Session continuity: pick your AI session back up on any device instead of starting over
- Bring your own model
- Desktop app, terminal UI and SDK
- Sub-agents that work in parallel under one supervisor
- Source code is public on GitHub
Feedback that would help me most
1. Which MCP servers did you connect first, and were the connection steps clear?
2. Which agent mode do you reach for, and what's missing?
3. Anything that broke or confused you in the first 10 minutes?
Replies
Hi Product Hunt! 👋 I'm Sayidi, a student from Bangladesh, and I built ZYRAXON AI on my own, with one laptop and no team.
Why I built it
Most AI tools answer questions. I wanted one that actually does the work: you give it a goal, and it plans the steps, picks the right tools, runs them, reads the results, fixes its own mistakes, and checks that the job is really done.
What's inside
- 888+ built-in tools across 63 categories
- A catalog of 8,202 MCP servers, each labeled with how far it was tested. Every server connects the way it needs to: nothing to do, a browser sign-in, an API key, or your own URL
- 11 agent modes (plan, build, explore, auto and more)
- Screen vision, so the agent can see what's on your screen
- Session continuity: pick your AI session back up on any device instead of starting over
- Bring your own model
- Desktop app, terminal UI and SDK
- Sub-agents that work in parallel under one supervisor
- Source code is public on GitHub
Feedback that would help me most
1. Which MCP servers did you connect first, and were the connection steps clear?
2. Which agent mode do you reach for, and what's missing?
3. Anything that broke or confused you in the first 10 minutes?
I read and reply to every comment. If you'd like to help build it, issues and pull requests are open: github.com/onelpawarai-X/ZYRAXON-AI
Thanks for taking a look! 🙏