Minotauris
Minotauris is your personal AI team for executing workflows
Most AI harnesses ask one agent to understand the goal, create a plan, use tools, review the result, and recover from mistakes.
Minotauris separates those responsibilities.
A Leader maintains direction.
Managers coordinate the work.
Workers execute.
The Assembly critiques plans before action.
The Windows beta can operate across the browser and desktop, run scheduled tasks, handle coding workflows, and support remote control.
The full demo is attached here:
I am testing whether structured role separation can make AI execution more reliable and less expensive than placing every responsibility inside one agent loop.
Minotauris is shifting toward a sharper wedge:
Repeatable browser and desktop SOPs → scheduled AI agent workflows.
A lot of work still happens across tabs, dashboards, spreadsheets, desktop apps, and internal tools. APIs do not cover everything, and traditional automation breaks when the workflow gets messy.
The idea is simple:
give Minotauris a recurring workflow, let agents run it, log what happened, and escalate when review is needed.
Curious who else is dealing with repetitive browser/desktop workflows that still require a human watching the screen.
-
1
I like the narrower direction.
One thing I'd keep testing is whether customers think they're buying automation or continuity. Most repetitive workflows already can be automated in pieces. The bigger problem is that someone still has to notice when they break, hand work off, or decide what happens next. That feels like a stronger category than simply "AI agents for browser workflows."
I’ve been building Minotauris, an automation and coding workspace based on coordinated AI agents.
The biggest thing I’ve learned: AI works better when it’s not treated like one giant prompt.
Most tools throw a model at a task and hope it figures everything out. That gets expensive fast, and quality drops when the task gets messy.
Minotauris is different because the work is structured.
One leader decides the direction. Managers break down the task. Workers execute specific parts across coding, browser work, files, and automation.
That makes the system more efficient:
fewer wasted tokens
less random tool use
better task handoff
lower cost
higher quality output
AI collaboration has been way more powerful than I expected, but only when the agents have clear roles instead of all acting like the same brain.
I’m still building and testing Minotauris, but this architecture feels like the right direction for real automation.
Would love feedback from other builders working with agents or automation.
Most AI products still treat serious work like a conversation.
That works for quick prompts.
But once the task becomes long-running, tool-heavy, or needs multiple steps, chat starts breaking down. You lose the plan. You lose visibility. Tool actions get buried. Approvals become awkward. And if the workflow fails halfway through, recovering the work is painful.
That is the problem I’m building around with Minotauris.
Minotauris is a local-first desktop app for running coordinated AI agent teams on a visual Team Canvas. Instead of hiding the work inside one thread, it shows the mission: plans, handoffs, todos, notes, tool actions, approvals, and outputs.
The direction is:
AI agents should be visible, not opaque
desktop automation needs approval gates
long-running work needs persistent state
local-first matters when agents touch real files and tools
multiple agents should coordinate like a team, not fight inside one prompt
Current beta is open, free, and BYOK.
Website: https://www.minotauris.app/
I’m still early, but I’m trying to figure out the clearest positioning:
Is the stronger angle “AI team canvas for desktop work” or “AI operating system for agent workflows”?
Comment