Workflows & tasks
| Feature | Description |
|---|---|
| Workflow library | Save a ComfyUI or Nerfstudio pipeline as a reusable template the whole team can run. Each workflow declares its parameters (with defaults and ranges) so anyone — or any AI agent — can submit a task without seeing the underlying graph. Learn more. |
| Task submission and tracking | Submit a workflow as a task, watch it through pending → queued → running → completed, and pull the output files when it finishes. Cancel any task that hasn’t reached a terminal state. Learn more. |
| Workflow snapshots | Every workflow save creates a snapshot you can preview and restore. Re-renders apply only to the parts that actually changed — no whole-project rebuilds. Learn more. |
Agents & compute
| Feature | Description |
|---|---|
| Team-owned agents | Connect a GPU machine — your own workstation, an on-prem server, a cloud VM — as an agent for your team. The dashboard shows which agents are online, their labels, and which tasks they’re handling. Learn more. |
| Label-based task routing | Tag each agent with labels (comfyui, nerfstudio, gpu-class, region, etc.) and require matching labels on a workflow so tasks land on the right hardware. |
| Comfy Cloud fallback compute | Connect a Comfy Cloud API key as an overflow target. When your team’s agents are busy or offline, tasks that need ComfyUI compute can spill to Comfy Cloud automatically — same workflow, hosted GPUs. Learn more. |
Container services
| Feature | Description |
|---|---|
| Deploy any HTTP service | Push a Docker image (vLLM, Ollama, a custom Flask app) to the team’s agents as a long-running service. The platform handles deploy, health checks, and traffic routing through a stable URL. Learn more. |
| Register an existing external service | Already running a model on your own infrastructure? Register the URL instead of deploying through Cyberun. The team’s agents make the outbound calls — Cyberun never makes direct outbound HTTP to a user-provided URL, so the network trust boundary stays on your side. MCP-callable just like a deployed service. Learn more. |
| Call container services from AI agents | Any MCP-aware AI client (Claude, Cursor, Codex CLI, etc.) can list and invoke the team’s container services through the Cloud MCP server. Whether the service is platform-deployed or externally registered is transparent to the AI — the same usage_prompt and call_container_service tool covers both. Learn more. |
Models & assets
| Feature | Description |
|---|---|
| Model library | Upload checkpoints, LoRAs, embeddings, and ControlNets once. The platform distributes them to eligible agents and caches them locally so workflows can reference any model by name without per-machine copying. Learn more. |
| Custom node distribution | Upload versioned ComfyUI node packages once. Agents pull and cache them on first need so every workflow runs against the same node versions across the team — no manual sync, no drift. Learn more. |
Notifications & events
| Feature | Description |
|---|---|
| In-app task notifications | When a task you submitted completes or fails, the dashboard surfaces it. No email, push, or polling to configure — open the dashboard, see what finished. Learn more. |
| Signed task webhooks | Subscribe an HTTPS endpoint to task events (task.completed and task.failed). Every delivery is HMAC-signed so your backend can verify the request really came from Cyberun before acting on it. Learn more. |
Identity & access
| Feature | Description |
|---|---|
| Teams and roles | Group users into teams with four role tiers (Owner / Admin / Member / Viewer). Permissions cover every resource — workflows, tasks, agents, container services — so a Viewer can browse without being a vector for accidental writes. Learn more. |
| Integration credentials | Issue sk- integration keys for scripts, CI, and AI agents. Each key is bound to a team at issue time; revoke at any time. No X-Team-ID header to manage. Learn more. |