llama.cpp appeared on Hacker News with a landing page for local AI software, drawing a score of 320 and 146 comments for a tool that runs frontier AI entirely on a user's own machine. The project describes itself as open-source, private and always local, with no API keys, no telemetry and no usage limits. Users can install the CLI with curl -LsSf https://llama.app/install.sh | sh, use Brew or Winget, or follow source-build instructions.

The page says llama.cpp can pair with a local coding agent by running llama serve, installing the pi-llama plugin and launching Pi. The setup is described as automatic local model discovery with no configuration and no API keys, while files stay on the machine and requests never leave it.

llama.cpp says it is optimized for hardware ranging from a laptop to a cluster, using the same binary, same models and hand-tuned kernels across GPU and CPU targets. Listed hardware includes Apple Silicon, M Ultra, RTX 5090, CPU, Jetson, H100, MI300, RTX 4090, A100, M Pro, M Max, DGX Spark, T4, Radeon RX, B200, Intel Arc and RTX 3090. The model catalog highlights Qwen 3.6, Gemma 4, GPT-OSS and Gemma 3, including features such as Dense and MoE variants, multimodal reasoning, agentic workflows, function calling, tool use, 140+ languages and up to 128K context.

llama.cpp appeared on Hacker News with a landing page for local AI software, drawing a score of 320 and 146 comments for a tool that runs frontier AI entirely on a user's own machine. The project describes itself as open-source, private and always local, with no API keys, no telemetry and no usage limits. Users can install the CLI with curl -LsSf https://llama.app/install.sh | sh, use Brew or Winget, or follow source-build instructions.

The page says llama.cpp can pair with a local coding agent by running llama serve, installing the pi-llama plugin and launching Pi. The setup is described as automatic local model discovery with no configuration and no API keys, while files stay on the machine and requests never leave it.

llama.cpp says it is optimized for hardware ranging from a laptop to a cluster, using the same binary, same models and hand-tuned kernels across GPU and CPU targets. Listed hardware includes Apple Silicon, M Ultra, RTX 5090, CPU, Jetson, H100, MI300, RTX 4090, A100, M Pro, M Max, DGX Spark, T4, Radeon RX, B200, Intel Arc and RTX 3090. The model catalog highlights Qwen 3.6, Gemma 4, GPT-OSS and Gemma 3, including features such as Dense and MoE variants, multimodal reasoning, agentic workflows, function calling, tool use, 140+ languages and up to 128K context.