Select llama.cpp GGUF on CPU Basic Spaces

Default extraction to the fine-tuned MiniCPM-V Hub checkpoint.

Remove dead chat code and obsolete planning docs.

Expand lab knowledge graph to 107 markers with curated YouTube videos.

Polish section 3 trace panel and Space startup reliability.

Add agent trace panel, vision extraction, and local Transformers backend.

Route ZeroGPU auto backend to Transformers

Fall back from Transformers to llama.cpp

Select extraction backend by CUDA availability

Switch extraction to PDF text only

Stabilize llama.cpp multimodal decoding

Increase llama.cpp ZeroGPU duration

Drop unsupported llama.cpp reasoning flag

Expose llama.cpp worker errors

Fix llama.cpp JSON extraction path

Make llama.cpp ZeroGPU backend default

Wire llama.cpp ZeroGPU badge path

feat: working Modal LoRA fine-tune + llama.cpp-on-ZeroGPU backend

Switch Space deployment to ZeroGPU

Add knowledge graph health report pipeline

Add local model bundle and report UI refinements

feat(extraction): offline llama-server backend + tolerant JSON parsing

feat(extraction): GBNF grammar + local MiniCPM-V (llama.cpp) backend + factory