Built Kivarro, an all-in-one local inference workbench. Looking for brutal feedback from people who actually run models locally.
这条记录涉及生成能力或端侧推理进展，适合跟踪模型效率、部署门槛和应用机会。
One app for chat. One app for model files. One script for llama.cpp flags. One dashboard for memory. One terminal for logs. One random note somewhere for benchmark results.
So Kivarro is my attempt at an all-in-one local inference app. Not a hosted service. Not a wrapper around a cloud API. A local-first desktop workbench for people running models on their own machines.
- cross-platform builds: Windows, Windows ARM64, macOS Intel, macOS Apple Silicon, Linux x64, Linux ARM64
I’m not claiming it is perfect. It is early. Builds are unsigned. The RAG part is currently a workbench, not automatic prompt injection. Agents are still a draft/control-plane area. The app is source-available under a non-commercial license.
What I want from this sub is feedback from people who actually run local models:
I built this. I want it to be useful. I’m looking for criticism before I build the next layer.