// the inferx lifecycle

    Model Lifecycle Loop

    A continuous loop — from upload to serving to eviction and back. No idle GPUs. No cold-start penalties.

    ↓
    ↓
    ↓
    ↓
    ↻ loop
    1

    Upload Model

    Push your model (HF, custom, fine-tuned)

    $ inferx model upload --name my-llm --source ./weights
    2

    Create GPU Snapshot

    Pre-warm into NVMe/RAM — instant recall

    $ inferx snapshot create --model my-llm --retain
    ✓ Snapshot ready — $0.10/hr retention
    3

    Instant On-Demand

    Traffic arrives → GPU loaded in <2s

    <2s
    4

    Evict from GPU

    Idle detected → freed in ~10ms

    ~10ms eviction
    5

    Restore in Seconds

    Traffic returns → back on GPU instantly

    → Traffic spike detected
    ✓ Model restored from snapshot — serving in seconds
    ↻ Loop continues...
    ↻

    Cycle repeats automatically — zero manual intervention.

    Ready to see it in action?

    Deploy your first model and watch the lifecycle loop handle everything.