AI Pulse · Under the Radar

Most AI work breaks down when you switch tools halfway through. One model has the research, another is better at writing, and a third might handle code or analysis. The usual handoff is a summary pasted into a new chat. That loses the useful parts, including how the brief changed, what was already tried, what failed, and which files or decisions matter.

Jiaojie is an open-source skill built to carry that fuller working context between AI models. It exports conversation state across HOT, WARM and COLD memory layers, rather than treating every handoff as a fresh start with a short recap. In practical terms, that could mean an employee can move a task to another model without asking it to repeat work, rediscover constraints, or make the same failed attempt again. The project was localised into six languages within a week of launch, and has 103 GitHub stars, growing at about 16 per day. That interest is not proof it works reliably at company scale, but it does point to a real frustration people are trying to solve.

For operators, the lesson is bigger than this one project. If your team uses more than one AI tool, context management is becoming part of the operating system. You need a clear record of the task, source material, decisions, outputs, and unresolved questions, not just a chat history trapped inside one vendor’s interface. This is the kind of thing we build into an AI command centre, so work can move between people and tools without losing its thread.