llama_context: n_ctx_seq (4096) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
UNSTUCK_BACKEND = offgrid
OFFGRID_GGUF_PATH = ~/github/unstuck/models/Qwen3-4B-Instruct-2507-Q4_K_M.gguf
generate() defined in module: unstuck.backend


=== OUTPUT FROM THE LOCAL MODEL ===
{
  "steps": [
    {
      "text": "Clear all items from the desk and sort them into 'keep', 'return', and 'discard' piles.",
      "minutes": 3
    },
    {
      "text": "Organize the 'keep' items by category (e.g., documents, supplies, devices) and place them in designated spots.",
      "minutes": 4
    },
    {
      "text": "Clean the desk surface with a cloth and wipe away dust; finally, place a fresh notepad and pen in a visible spot.",
      "minutes": 3
    }
  ]
}