# Probe: explicit CPU backend with build disabled

Command:
node .opencode/specs/system-spec-kit/026-graph-and-context-optimization/014-local-embeddings-migration/041-llama-cpp-metal-investigation/scratch/probe-gpulayers-zero.mjs cpu

Exit: 1

Output:
```text
[node-llama-cpp] A prebuilt binary was not found
file:///Users/michelkerkmeester/MEGA/Development/Code_Environment/Public/.opencode/skills/system-spec-kit/mcp_server/node_modules/node-llama-cpp/dist/bindings/getLlama.js:279
        throw new NoBinaryFoundError();
              ^

NoBinaryFoundError: NoBinaryFoundError
    at getLlamaForOptions (file:///Users/michelkerkmeester/MEGA/Development/Code_Environment/Public/.opencode/skills/system-spec-kit/mcp_server/node_modules/node-llama-cpp/dist/bindings/getLlama.js:279:15)
    at async file:///Users/michelkerkmeester/MEGA/Development/Code_Environment/Public/.opencode/specs/system-spec-kit/026-graph-and-context-optimization/014-local-embeddings-migration/041-llama-cpp-metal-investigation/scratch/probe-gpulayers-zero.mjs:33:15

Node.js v25.6.1
```

Interpretation:
`getLlama({ gpu: false, build: "never" })` cannot use a CPU-only prebuilt from the current install. A previous unguarded CPU probe attempted the package's source-build/download path and failed because network access is disabled; the saved scratch script now uses `build: "never"` to avoid that side effect.
