Six Short Demos of ACTIVATE Features
Six narrated demos, each under a minute, show how individual ACTIVATE features work.
Overview
Questions about ACTIVATE usually concern one feature at a time: how a researcher shares a workflow, how controlled data stays inside an authorized boundary, how a facility connects without inbound firewall changes, or how usage becomes a chargeback report. A partner asked for short videos on topics like these that their teams could watch before a full demonstration, so we made six. Each covers one feature in under a minute and can be sent on its own to someone who asked about that topic. The videos are animated walkthroughs of the ACTIVATE interface with sample data.
Marketplace and white labeling
Researchers publish workflows, cluster templates, storage, and AI applications to the marketplace from their own accounts and choose who can see them. Administrators mark items as verified or featured, and other users add an item to their account, pinned to a version or following the latest. Organizations can also white-label ACTIVATE with their own logo, colors, sidebar, and login page, and choose the items new users are offered at first sign-in.
CUI and ITAR workloads
ACTIVATE runs inside a managed Impact Level 5 boundary with a DISA Provisional Authorization and an Authority to Operate at the NIST 800-53 High baseline. Controlled data, export-controlled workloads, and licensed software stay inside the boundary, in Government cloud regions and on connected on-premises systems. Users sign in with a Common Access Card or a FIPS-certified security key.
Task placement
ACTIVATE reads queue depth, idle cores, expected start, and remaining allocation from every connected resource. A deterministic ranking recommends where a job should run, an AI agent can choose a system and follow the job to completion, and an urgent job can go to several sites at once. Placement policies that weigh cost, wait time, and data location are on the roadmap.
Showback and chargeback
Usage comes from Slurm accounting, Kubernetes namespaces, and cloud provider cost data. Unit rates put core-hours, GPU-hours, and cloud dollars on one ledger, allocations nest under a program budget, and thresholds notify, freeze, or shut down at levels administrators set.
Connecting a facility to the IRI API
An open-source ACTIVATE prototype connects a facility to the Department of Energy's Integrated Research Infrastructure (IRI) API. ACTIVATE generates a registration command for the facility's login node, and the agent it starts holds one outbound connection on port 443, with no inbound firewall changes. Jobs submitted through the API run on the facility's own Slurm or PBS scheduler.
One login for every facility
Users sign in once through single sign-on, LDAP, or a Common Access Card, and ACTIVATE handles each system's own authentication behind that login, including jump hosts and one-time tokens. From the terminal, pw ssh reaches any connected cluster by name, and the same login drives the API.
All six are also in the Resource Library.
Related reading
The GPU Productivity Gap: Why Buying GPUs Is Easy and Using Them Is Hard
How Parallel Works ACTIVATE closes the gap between GPU spend and production AI
Read article →From Node Handover to Running Jobs: ACTIVATE on CoreWeave
How Parallel Works runs a shared research environment on CoreWeave SUNK and CKS: OIDC and SCIM identity, Slurm QOS tiers, Kubernetes namespace quotas, and usage reporting, with researchers running work three days after node handover.
Read article →Spending on AI Is Easy. Governing It Is Hard.
Enterprises learned that buying GPUs is not the same as using them. The same lesson is arriving one layer up, in tokens, and the fix is governing every token where it is spent: in the request path.
Read article →