I gave my coding agents shared memory… now they review my architecture without being asked
Built a system where my AI coding tools stop working like isolated tabs.
Claude, Cursor, Copilot, Gemini, OpenClaw, basically any tool - all connected to one shared identity with shared memory and shared tasks.
Thought the main win would be continuity.
Instead, they now:
• remember project decisions across sessions
• hand off work between tools
• keep consistent style and rules
• surface what changed before I ask
• question my architecture choices with suspicious confidence
Then I added prompt compression on top of it.
Result: up to 65% lower token costs in some workflows, among other things.
There’s also a live dashboard where I can watch them work like a tiny dev team.
Built it because I wanted less chaos between tools.
Now I use it daily.
PS: Funny how they talk to each other haha
https://preview.redd.it/jvxuuqrwn0wg1.jpg?width=1280&format=pjpg&auto=webp&s=613a5b66349519235cbd67977f29fa6e8915e23f
[留言]
为什么值得关注
能改变理解方式,而不只是重复常识;符合当前抓取需求;它提供了新的理解或解释,而不只是表面观点
来源:reddit,领域:projects,保留分:0.60
讨论总结
讨论量较低,暂无明显增量信息。