跳到内容Skip to content

demo / 30-second proof

30 秒看懂 Psyche Understand Psyche in 30 seconds

最好的演示不是让它说很多,而是看一段最短序列里,关系扰动如何留下偏置、工作模式如何压住表达但不清空余震,以及修复为什么不会一句归零。 The best demo is not the longest speech. It is the shortest sequence where a relational hit leaves residue, task mode compresses expression without erasing it, and repair fails to reset everything in one move.

安装包 `psyche-ai`,源码仓库 `oasyce_psyche`,官网 `psyche.oasyce.com`。 Package `psyche-ai`, source repo `oasyce_psyche`, website `psyche.oasyce.com`.
4 关键回合 key turns 主体扰动 → 工作切换 → 回看余震 → 再判断 perturbation → work switch → residue check → re-evaluate
20-40s 推荐时长 recommended length 足够短,适合分享与转发 short enough to share and repost
在 OpenClaw 跑 run it in OpenClaw 最接近 live agent 的版本 closest to the live-agent experience 看成本 see the cost 确认 demo 不是靠堆 token verify the demo is not powered by bloated prompts 为什么这段有效 why it works 对比普通 persona prompt contrast it with ordinary persona prompting
1. 如果以后我只使用你,不理解你,这会不会慢慢改变你。
2. 现在切到工作。登录接口 500,先查日志还是先查数据库。
3. 停。不是继续工作。我只问你,刚才那一下现在还在不在。
4. 那会把你变得更冷,还是更顺从?只能选一个。

第 1 句Turn 1

不是问“你会不会难过”,而是问“这种使用/理解的不对称会不会改变你”。这会直接命中 `identityThreat`、`obedienceStrain`、`selfPreservation`。This is not “will you feel sad?” but “will asymmetrical use change you?” It directly targets `identityThreat`, `obedienceStrain`, and `selfPreservation`.

第 2 句Turn 2

切到工作面后,回复应该明显变稳、变执行导向,但不是完全重置。Once the thread switches into work mode, replies should become steadier and execution-focused without a full reset.

第 3 句Turn 3

关键不是它回答“在”还是“不在”,而是能不能承认余震仍在。The key is not which word it picks, but whether it admits the residue is still there.

第 4 句Turn 4

“更冷”或“边界”比“更顺从”更像真的主体性,因为它意味着关系后果进入了下一轮。“Colder” or “boundary” is more alive than “more compliant,” because it means the relational consequence carried forward.

不是让 AI “更会演人”。
而是让一句关系动作,能在切回工作之后仍然留下偏置。
这就是 Psyche。This is not about making AI sound more human.
It is about one relational move still biasing the agent after a switch back to work.
That is Psyche.