Repository navigation

Add managed llama.cpp engine with Windows CUDA support - #116

Closed

nv-pgoode wants to merge 23 commits into

NVIDIA:developfrom

nv-pgoode:codex/llamacpp-core-minimal

Closed

Add managed llama.cpp engine with Windows CUDA support#116
nv-pgoode wants to merge 23 commits into
NVIDIA:developfrom
nv-pgoode:codex/llamacpp-core-minimal

Commits on Sep 22, 2026
- Add managed llama.cpp as a third engine across the services
  
  committed
  a48d357
- Walk ancestors in the Unix owned-path guard
  
  pgoode41
  and
  claude
  committed
  85fb8b4
- Add llama.cpp to the desktop
  
  committed
  ffc7d5d
- Document llama.cpp and teach the tooling about it
  
  committed
  7a178ae
- Focus workload display and repair engine event consumers
  pgoode41
  committed
  56e837d
- Defer explicit workload cancellation from the llama.cpp integration
  pgoode41
  committed
  dd958ba
- Include embedded engine manifests in the service build fingerprint
  
  pgoode41
  and
  claude
  committed
  66eb109
- Repair llama.cpp integration findings from the whole-engine assessment
  
  pgoode41
  and
  claude
  committed
  21873ca
- Route llama.cpp like the other engines and load models on demand
  
  pgoode41
  and
  claude
  committed
  931647a
- Finish the routing wording in the desktop docs
  
  pgoode41
  and
  claude
  committed
  172c8bd
- Preserve live engine status across delayed startup snapshots
  pgoode41
  committed
  f305c3b
- Install the official CUDA llama.cpp build on Windows x64 NVIDIA GPUs
  pgoode41
  committed
  26271a0
Commits on Sep 23, 2026
- test: respect Unix cache environment key case
  pgoode41
  committed
  81624e8
Commits on Sep 25, 2026
- Merge develop into the managed llama.cpp integration
  
  pgoode41
  and
  claude
  committed
  8d88a57
- Verify the llama.cpp accelerator everywhere and gate CUDA on Linux
  
  pgoode41
  and
  claude
  committed
  fefad2c
- Classify Metal and ROCm devices and surface stall reasons on the gated paths
  
  pgoode41
  and
  claude
  committed
  a4261e4
Commits on Sep 26, 2026
- Count cache growth and child I/O as transfer progress
  
  pgoode41
  and
  claude
  committed
  a7391b0
Commits on Sep 27, 2026
- Show a llama.cpp install in progress as managed, not external
  
  pgoode41
  and
  claude
  committed
  8d8c8ff
- Download llama.cpp archives and installers into the install stage
  
  pgoode41
  and
  claude
  committed
  fb759bd
- Pin per-engine attribution in the engine:models-changed snapshot
  
  pgoode41
  and
  claude
  committed
  cf22669
- Hold a busy owner against new admissions instead of queuing into it
  
  pgoode41
  and
  claude
  committed
  6a51524
Commits on Sep 28, 2026
- Retry the llama.cpp runtime promotion while Windows still holds the files
  
  pgoode41
  and
  claude
  committed
  9c8632c
- Report a fully streamed response as completed when the client closes early
  
  pgoode41
  and
  claude
  committed
  89ab11a

Commits

Commits on Sep 22, 2026

Commits on Sep 23, 2026

Commits on Sep 25, 2026

Commits on Sep 26, 2026

Commits on Sep 27, 2026

Commits on Sep 28, 2026