Launched this week
MaxHermes by Minimax
AI agent that builds skills from every task you give it
106 followers
AI agent that builds skills from every task you give it
106 followers
MaxHermes is a cloud sandbox AI agent that autonomously extracts reusable skills from completed tasks and improves itself across sessions. For knowledge workers and enterprise teams who want an AI assistant that compounds with use.
MaxHermes is a cloud sandbox AI agent from MiniMax that autonomously extracts reusable skills from completed tasks and applies them in future sessions.
Problem → Solution: The statelesness of AI agents is a design problem that the industry has mostly ignored. You do the work of teaching the agent your context, it performs, the session ends, and nothing carries forward.
MaxHermes introduces a learning loop at the execution level. Every complex task triggers automatic skill extraction. Those skill documents persist, get refined through feedback, and stack over time. The agent that handles your tasks in month three is meaningfully more capable than the one you started with.
What makes it different: Built on MiniMax M2.7, it achieves this without requiring any manual capability setup or server configuration. The evolution is autonomous.
Key features:
Learning loop that auto-generates skills post-task
Cross-session persistent memory
7x24 cloud availability with no infrastructure on your end
Integrations with Feishu, DingTalk, and Enterprise WeChat out of the box
Benefits:
Your AI assistant improves from real usage, not from prompt tuning
Zero setup cost for enterprise teams already in Chinese productivity ecosystems
Parallel task execution means it handles multiple workstreams without queuing
The compounding skill model is the bet worth watching here. Whether it holds up at scale and across diverse task types is the open question at launch.
P.S. I hunt the latest and greatest launches in tech, SaaS and AI, follow to be notified → @rohanrecommends
Open Wearables
cloud sandbox approach makes sense for this. how does the skill library work across team members? wondering if there's a way to share learned skills between agents or if each instance builds its own knowledge base.
Congratulations