Why Cremind?
Five advantages that set Cremind apart from other agent platforms — model groups, event-driven design, Agent Skills, Cremind Connect, and self-hosting.
Plenty of tools can call an LLM in a loop. Cremind's design choices are about making a personal agent that is cheaper to run, reacts in real time, and stays under your control. Cremind is early (version 0.0.1) and community-driven, so these are design principles in active development — not a finished, polished platform. Here is what makes it different.
1. Two model groups, one focused agent
Cremind splits work across two model groups. A high group does the heavy reasoning and planning; a low group runs individual tool calls and routine steps. Because you don't pay frontier-model rates for every small action, the result is lower token cost without giving up planning quality.
2. Real-time and event-driven
Most agents poll: they wake on a timer, check for changes, and go back to sleep. Cremind is event-driven instead. Skills can push real-world events straight into the agent over a live connection the moment they happen — no cron, no polling loop — so the agent reacts with sub-second responsiveness when something actually changes.
3. Agent Skills with event listeners
A Skill in Cremind is just a SKILL.md file plus a few scripts — easy to read, easy to write, easy to share. What makes them powerful is that a skill can also ship a long-running listener, so a skill is not only a set of commands the agent can call but also a source of the events it reacts to.
4. Cremind Connect for integrations
Wiring an agent into Gmail, Calendar, or Jira normally means standing up your own cloud project and juggling OAuth credentials. Cremind Connect is an open-source OAuth broker that links those services without that setup, and it is a separate sub-product with its own site.
Cremind Connect
What Connect does and how it links your accounts. Full details at connect.cremind.io.
5. You own it
Cremind runs on hardware you control. Your keys, conversations, and indexed documents stay on your machine or server. Profiles keep multiple assistants isolated on one install, and an optional sandboxed Docker mode adds stronger separation between the agent and your host. Being open source under the MIT license, you can read every line, change it, and run it however you like.
Want to see all of this in action? Head to the Quickstart and you'll be chatting with your own agent in a few minutes.