Persistent Workspaces

Reserved GPU. Burst When Available.

Stop racing the pod market. Your Qiln workspace gets a dedicated primary GPU allocation by plan — a full GPU or GPU slice (MIG) depending on the workspace. Keep the same environment, folders, routes, and models; add burst GPUs when the pool has room.

  • Dedicated GPU or GPU slice by plan
  • Optional burst capacity when available
  • No shared-VRAM surprises from other users
Feature screenshot
Persistent Storage

Real Folders. Private Model Vaults.

Your ComfyUI setup should not vanish with the pod. Keep checkpoints, LoRAs, VAEs, ControlNets, workflows, custom nodes, inputs, and outputs in persistent workspace folders. Snapshot before risky node changes; roll back when dependencies break.

  • Private model vaults with direct file access
  • Persistent outputs, workflows, and custom nodes
  • ZFS-backed snapshots, clones, and rollbacks
Feature screenshot
workspace blueprints

Blueprints for persistent AI workspaces.

comfyui-workspace.yaml 
name: comfyui-workspace
display_name: Persistent ComfyUI Workspace
image_alias: ubuntu:24.04-cuda

apps:
  - name: comfyui
    route: https://{{ workspace.slug }}.qiln.com
    command: /opt/comfyui/start.sh
  - name: vscode
    route: https://{{ workspace.slug }}-code.qiln.com

gpu:
  primary: reserved
  burst: true

volumes:
  - name: model-vault
    type: private_vault
    mount_path: /opt/comfyui/models
  - name: outputs
    type: persistent_folder
    mount_path: /opt/comfyui/output
  - name: custom-nodes
    type: persistent_folder
    mount_path: /opt/comfyui/custom_nodes

snapshots:
  before_updates: true
  retain: last_10

endpoints:
  - name: workflow-api
    app: fastapi-wrapper
    route: https://{{ workspace.slug }}-api.qiln.com

Workspace, not pod

ComfyUI and VS Code get stable routes inside one durable workspace instead of another throwaway GPU session.

Reserved primary GPU

The workspace has primary GPU capacity reserved by plan, with burst capacity available when the pool has room.

Real folders and rollback

Models, outputs, and custom nodes mount as persistent folders with snapshots before risky updates.

Path to endpoints

When a visual workflow becomes useful, wrap it behind a stable API route instead of rebuilding somewhere else.

A Qiln blueprint describes the workspace a user opens — apps, folders, GPU capacity, routes, snapshots, and endpoint paths.