We just tested DeepSeek Harness, and it's more flexible than we expected. We initially assumed it was another Claude Code, but Harness can switch between multiple models on the fly and connect to vision-capable models, letting DeepSeek analyze images. Its team describes the goal as "temporal" and "spatial composability": tools can be installed, adjusted and removed dynamically without breaking the system. This would be a platform built for endless mods without constant crashes.
AI 导读
我们刚测试了 DeepSeek Harness，它比我们预期的更灵活。我们最初以为它又是一个 Claude Code，但 Harness 可以在多个模型之间即时切换，并连接支持视觉的模型，让 DeepSeek 分析图像。其团队将目标描述为“时间性”和“空间性可组合”：工具可以动态安装、调整和移除，而不会破坏系统。这将是一个为无限扩展而构建、不会频繁崩溃的平台。
We just tested DeepSeek Harness, and it's more flexible than we expected. We initially assumed it was another Claude Code, but Harness can switch between multiple models on the fly and connect to vision-capable models, letting DeepSeek analyze images. Its team describes the goal as "temporal" and "spatial composability": tools can be installed, adjusted and removed dynamically without breaking the system. This would be a platform built for endless mods without constant crashes.
来源：X.PIN· x.com