Early access

SciFlow

Self-service GPU compute for private research clusters.

SciFlow gives labs and platform teams a product interface for Kubernetes GPU clusters: templates, admission, queues, workspaces, storage, images, logs, metrics, and API access without exposing Kubernetes to researchers.

Private cluster control
Quota-aware queueing
Templates instead of YAML
Live GPU utilization
Image and storage reuse

Request early access Try the live demo

Launch instance

PyTorch + Jupyter

Starts now

H100

A100

L40S

admission preview

policy ok

Templatepytorch:2.4 + jupyter

Org chargedVision Lab

GPU shape1 × H100 (1/1)

Auto-stop12h

CPU

24 cores

VRAM

80 GB

ETA

now

Templates

34 reusable

Quota

6 / 8 H100

Images

142 saved

The problem

Research compute should not depend on YAML, IM messages, and manual kubectl

Researchers lose time translating experiments into cluster configuration. Admins lose time policing fair access and reconstructing what happened after a run fails. SciFlow replaces that patchwork with one product surface.

Cartoon illustration of manual GPU access with scattered notes and chat messages

Problems

Kubernetes is not a research UIMost researchers want a notebook, SSH shell, or dev server, not resource manifests, PVC details, and service wiring.
GPU access is hard to share fairlyWithout a quota-aware admission layer, the loudest user wins and queues become IM threads.
Templates drift into copy-pasteEvery lab repeats setup scripts for CUDA, Conda, Jupyter, SSH keys, storage mounts, and startup commands.
Utilization and events are scatteredGPU usage, queue decisions, instance events, and failure reasons live across kubectl, metrics dashboards, and logs.

Cartoon illustration of an organized GPU research workspace dashboard

SciFlow solutions

Multi-tenant policyPlatform admins define orgs, quotas, member limits, and admission rules so GPU access follows policy.
Researcher-friendly interfaceResearchers launch templates, notebooks, terminals, apps, storage, and saved images without writing YAML.
ObservabilityGPU utilization, queue state, logs, events, and failure reasons are visible from the product surface.

Governance

Clear responsibility from platform to lab members

SciFlow keeps cluster-wide operations, organization policy, and day-to-day research access separate.

Platform admin: Owns tenants, integrations, GPU inventory, platform settings, pricing, and operational logs.
Org admin: Manages members, invitations, org GPU quotas, member limits, and org storage policy.
Org member: Launches templates, runs instances and jobs, mounts storage, saves images, and tracks usage.

How it works

From request to running workspace

The workflow stays short for researchers while policy and capacity checks run behind it.

01Choose

Select environment

Org
Template
GPU shape
Auto-stop

02Admit

Check policy

Membership
Quota
Capacity
Queue

03Work

Open workspace

Jupyter
SSH
App links
Storage

Platform capabilities

What the console actually gives each team

The product is organized around the screens researchers, org admins, and platform admins use every day.

Console surface

Template catalog and launch

Image familiesOptional appsAdmission preview

Screen

Launch

Console surface

Instances and jobs

Queued requestsApp linksStart, stop, requeue

Workloads

Live

Console surface

Storage and images

Workspace filesOrg storageImage commits

Reuse

Save

Console surface

Org administration

MembersInvitationsGPU quota ledgers

Scope

Lab

Console surface

Platform administration

OrganizationsGPU shapesRegistry and auth

Scope

Cluster

Console surface

Reporting and support

GPU-hoursGrafana panelsSupport threads

State

Trace

Early access

Bring self-service GPU compute to your cluster

Launch workspaces, enforce quotas, and expose runtime state without making researchers operate Kubernetes.

Request early access Review capabilities