This means you can build AI-powered validation, analysis, and decision steps directly into your workflow — without waiting for a purpose-built agent node to be added for your use case.

Release Highlights​

What is it?​

The Edison Node is a new workflow node that calls Edison (Rescale’s AI assistant) with a prompt you define. You control what the agent does by writing instructions in plain language, selecting which agent to use, and optionally attaching Connectors and Agentic Projects for additional context.

Why does it matter?​

Before the Edison Node, adding AI to a workflow meant using one of the pre-built agent nodes — each designed for a specific task. If your use case didn’t fit an existing agent, you were stuck. The Edison Node removes that constraint.

You can now:

• Provide custom instructions to any Edison agent
• Give agents access to specific Connectors (data sources, tools) relevant to your workflow
• Attach Agentic Projects that tailor agent behavior to your domain or team’s needs
• Combine the Edison Node with Decision nodes to route workflows based on AI reasoning

How is it used?​

Any user with access to Workflows can add an Edison Node from the workflow canvas. It’s available alongside the existing agent nodes (Input Validator, Deck Validator, Troubleshooter) in the node toolbar.

Getting Started​

• Open the Workflows page and create a new workflow or edit an existing one.
• From the node toolbar, drag an Edison node onto the canvas
• Configure the node:

Agent — Choose a specific agent or leave on Auto-select to let Edison pick the best one for your prompt.

• Prompt — Write the instructions for what you want Edison to do at this step.
• Connectors (optional) — Attach data sources or tools the agent should have access to.
• Agentic Projects (optional) — Select a project that shapes the agent’s behavior for your use case.
• Connect the Edison Node to other nodes in your workflow (e.g., after a Job node to analyze results, or before a Decision node to drive conditional branching).
• Click Save, then Run Workflow to execute.

For more on building workflows, see the Workflows documentation.

Release Highlights​

• What is it? A new capability of the Rescale MCP server that gives your AI assistant access to Rescale agents through Edison. Your assistant sends the question to Edison, and Edison decides which agents to use and combines their work into one answer, the same way it does inside Rescale.
• Why does it matter? Until now, Rescale agents were only available inside the Rescale platform. Engineers who plan, write, and code in another assistant could not leverage their capability. Now your assistant can combine an agent’s answer with your attached files and other connected tools. For example, it can ask Edison why a run failed and then edit the input deck you attached.
• How is it used? Active users of an Agentic Engineering workspace can use it, signed in with their own Rescale account. Rescale checks your access on every request. Edison and its agents see only what you can see in the workspace you chose at sign-in.

You can ask questions like:

• “Ask Rescale to troubleshoot job [JOB_ID] and suggest a fix.”
• “Have Rescale check this input file before I submit it.”
• “Ask Rescale what our standard mesh settings are for external aero, with sources.”
• “Use Rescale to find the slowest runs from last month’s crash study.”

Getting Started​

• Connect Rescale to your assistant. In Claude, open Customize → Connectors → Browse connectors and search for Rescale. For ChatGPT, Cursor, Codex, and other clients, add the Rescale MCP server by URL.
• Sign in with your Rescale account, choose your Agentic Engineering workspace, and approve access.
• Ask Rescale your question or describe the task. Edison decides which agents to use.

Agents need a sign-in with your Rescale account through SSO or password login. Connections set up with an API key get the standard job and file tools, but not Edison or agents. If your workspace was upgraded to Agentic Engineering after you connected, disconnect and reconnect to Rescale.

Known Limitations​

• Answers come back as text. Plots, tables, and files an agent produces are summarized rather than shown.
• If Edison needs you to pick an option partway through, the request stops. Continue in Edison in the Rescale platform, or rephrase the request with the choice included.

Edison and agent requests use AI Credits. For setup details and the agents included with Agentic Engineering, see Rescale MCP Server – Getting Started.

Release Highlights​

• What is it? A Rescale-hosted connector, built on the Model Context Protocol (MCP), that you install from Claude’s directory. You don’t need a server URL or an API key.
• Why does it matter? Engineers already plan and write in Claude. With the connector, Claude can work with Rescale alongside your attached files and other connected tools. For example, it can read an input deck you attached, set up the job on Rescale, and summarize the results when the job finishes.
• How is it used? Any Rescale user with MCP access can connect. Claude works in the workspace you choose at sign-in, with your own Rescale permissions, and every action is recorded as yours. Claude asks for your approval before the connector submits, stops, or deletes anything.
• What does each plan include? On every plan, including HPC Foundations, Claude can work with your jobs, files, and workstations and look up the software and hardware available to you. Advanced Modeling & Simulation adds workflows. Agentic Engineering adds Edison and its agents. Advanced Modeling & Simulation includes Edison in the Rescale platform, but using Edison and agents from Claude requires Agentic Engineering.

You can ask questions like:

• “Show my recently completed Rescale jobs and summarize the latest result.”
• “Job [JOB_ID] failed. Read the logs and explain what went wrong.”
• “Which versions of Simcenter STAR-CCM+ can I run, and on which hardware?”
• “Here’s my input deck. Set up an Ansys Fluent job on 64 cores and show me the settings before submitting.”
• “Ask Rescale to troubleshoot job [JOB_ID] and suggest a fix.” (Agentic Engineering)

Getting Started​

• In Claude, open Customize → Connectors → Browse connectors.
• Search for Rescale and click Connect.
• Sign in to Rescale, choose your workspace, and approve access.
• Start with a read-only request, such as “Show my recently completed Rescale jobs.”

On Claude Team and Enterprise plans, an organization owner may need to add Rescale before members can connect.

If your workspace has Agentic Engineering, ask Rescale Edison your question or describe the task. Edison will drive to a result and decide which sub agents to leverage. If your workspace was upgraded after you connected, disconnect and reconnect Rescale so Edison and its agents are available.

The directory listing connects to Rescale’s US platform. If your account is on the EU, Korea, or Japan platform, add Rescale to Claude as a custom connector with your region’s server URL. The connector isn’t available on the ITAR or FedRAMP platforms.

Submitting jobs or starting workstations through Claude uses compute at your normal rates. For setup details, regional URLs, and troubleshooting, see Rescale MCP Server: Getting Started.

Highlights​

• Private beta, enabled by your account team: Windows Server 2025 isn’t a self-service toggle yet. Contact your Rescale account team to have it enabled for your workspace.
• A newer Windows option on AWS and Azure: Run jobs and workstations on Windows Server 2025, unlocking software versions that no longer support Windows Server 2019.
• No disruption to current workflows: Windows Server 2019 remains fully available alongside 2025. This is an addition, not a cutover.
• Validated provisioning on both clouds: Rescale has tested cluster bootstrap and provisioning end-to-end on Windows Server 2025 for AWS- and Azure-hosted batch jobs and workstations.
• Priority support for blocked workflows: If your software has already dropped Windows Server 2019 support, your account team can help migrate your workflow to Windows Server 2025 this quarter.

Availability​

Windows Server 2025 is rolling out as a private beta. The software versions available now are Bring Your Own Windows Software Workstation (Windows), Custom (Windows Server 2025) and Bring Your Own Windows Software, Custom (Windows Server 2025).

Get Started​

Windows Server 2025 is in private beta and isn’t yet a self-service toggle. To run your jobs or workstations on it, contact your Rescale account team to have it enabled for your workspace.

Once enabled, select Windows Server 2025 the same way you’d select any other OS image when configuring your job or workstation. No other setup is required.

If your team is currently blocked because your software has dropped support for Windows Server 2019, let your account team know. We’re prioritizing migration of affected workflows this quarter.

Release Highlights​

An Agentic Project is a container for everything Edison needs to know about one engineering project:

• Objective: what the project is trying to achieve, in your words
• Linked jobs and workflows: golden jobs to duplicate, baselines to compare against, and saved workflow templates to run time and time again
• Linked files and folders: standards, reports, and requirements uploaded to Rescale or from connected knowledge bases, e.g. SharePoint, Amazon S3 and Azure Blob
• Skills: reusable instructions and scripts agents apply, such as checking input files against your standards
• Learnings: tribal knowledge, rules of thumb, and lessons Edison applies when they’re relevant, suggested by Edison or added by you

Why we built it

• No more re-explaining. The context loads into every chat the project is attached to, whether you come back tomorrow or in three months.
• Knowledge stays with the project. Standards, golden jobs and lessons learned live on the platform, not in one engineer’s head, so they stay when someone changes teams.
• Corrections become team rules. Tell Edison “always validate input files before submitting jobs” and it offers to save that as a learning for everyone who uses the project.
• Answers you can check. When Edison answers from a linked document, it cites the source.

An Agentic Project holds a project’s report template, golden jobs, standards, requirements and learnings in one place.

Getting Started​

• Go to Agentic Projects and create a project, or ask Edison to do it. Give it a name and say what you’re trying to achieve; that’s all it takes.
• Link what matters: golden jobs, standards and requirements files, a report template. You can do it on the project page or ask Edison, for example “Find my file in SharePoint and link it to this project.”
• In Work with Edison, attach the project to your chat. Edison and the agents it calls now work with the project’s context.
• Keep working. When Edison spots something worth keeping, it proposes a learning or a link; accept the ones you want saved.

Agentic Projects requires Agentic Engineering enabled on your workspace. Contact your workspace administrator or your Rescale account team to enable.

For more information see Agentic Projects.

Good to Know​

• Private until you share it. New projects are visible only to you. Share to the workspace when it’s ready; teammates can use a shared project, and only the owner can edit it.
• One project per chat. A new chat starts with no project attached.
• Linked, not copied. A project links up to 10 jobs and 15 individual files and folders. Edison reads linked files as the person asking, so no one sees content they couldn’t already open.
• You decide what’s saved. Learnings and links Edison suggests are saved only when you accept them.

Before this, checking resource usage meant logging into a vendor dashboard or SSHing into the cluster. Now it's on the job's detail page automatically, with no setup required.

Cluster Metrics only collects data going forward, so a job that completed before this feature shipped won't show any metrics.

See the Cluster Metrics documentation for details.

What’s Changed​

Edison opens from the left navigation. The small sparkle icon in the top-right corner has been removed. Look for Work with Edison in the left navigation instead. It’s the same sparkle icon, now with a label, in the same place on every page.

Chats. You can now keep several chats with Edison and return to any of them. Within a chat, Edison remembers what was said and what it looked up earlier, so you can continue days later without starting over. Click New chat to start a fresh one. Edison titles each chat from your first message. You can rename or delete chats from the chat list, and deleting one asks you to confirm first. Chats are private to you, stay in their workspace, and are deleted automatically after six months without activity.

Nothing else about your account changes. Your permissions, workspace settings, and AI Credits are the same.

Highlights​

• Always in reach: Rescale Edison sits in the left navigation on every page.
• Chats and Chat Memory. Users can keep several long-running chats, return to any of them, and rename or delete them. New chat replaces Clear.
• Save repeat processes: As you work with Edison to run your processes, Edison remembers your steps and can save your processes as workflows so it is easier to run in the future.
• Answers from your own data: Ask about jobs, files, workflows, results, or spending in plain language. Edison queries your Rescale account directly and charts the result where that helps.
• Within your permissions: Edison only reads what you could open yourself. Administrators get organization-wide or workspace-wide answers, depending on their role.

Work with Edison in the left navigation, with the panel open showing your chats, most recent first.

Get Started​

• Sign in to the Rescale platform.
• Click Work with Edison in the left navigation.
• Type your question and press Enter.
• Open any other page. Edison stays open beside it.

Edison needs Advanced Modeling & Simulation or Agentic Engineering enabled on your workspace. If Work with Edison shows as unavailable, contact your workspace administrator or your Rescale account team.

For complete documentation, see Rescale Edison.

Release Highlights​

What is it?

• Node-Level Failure Controls — Previously, the “continue on failure” setting applied to the entire workflow. Now you control it per node. Under Node Properties in each node’s configuration panel, toggle Continue workflow execution if this node fails to let independent branches keep running when that node fails. Only the failed node’s downstream dependents are blocked — everything else keeps going.

To apply the setting across all nodes at once, open Workflow Settings and click Set For All Nodes next to Continue on Failure. You can then override individual nodes as needed.

Note: Decision node failures still halt the workflow, since the system cannot determine which path to take without the decision result.

• Concurrent Node Limits — You can now cap how many nodes run at the same time in a workflow. In Workflow Settings, enable Concurrent Nodes and set the maximum with the numeric stepper. When the limit is reached, additional ready nodes wait until a running node completes.

This is useful when your workflow runs commercial solver software with a limited number of licenses — set the concurrency limit to match your available license count so the workflow never consumes more licenses than you have.

Set the max number of concurrently active nodes in Workflow Settings

• Retry Failed Workflow Runs — When a workflow run has failed nodes, click Retry on the run to fix what went wrong and re-run only what needs to re-run. Clicking Retry opens the workflow template editor with a color-coded overlay showing the status of each node:

Green: Completed, can be reused — succeeded and unchanged, will be skipped

• Red: Failed, will retry — failed or never ran, will re-execute
• Orange: Changed or downstream of a change, will re-run — configuration was modified or depends on a node that changed
• Gray: Did not run, will run — blocked by an upstream failure, will execute for the first time

Fix the configuration that caused the failure directly in the template, then click Retry Workflow to launch the continuation run. Nodes that succeeded and haven’t changed are skipped — their outputs are reused from the original run. The retried run links back to the original for traceability.

Review which nodes will re-run and which will be reused before confirming the retry

Why did we build it?

Teams running large, multi-branch workflows — dozens of nodes with hundreds of monthly runs — told us they needed three things: the workflow should compute everything it can when one node fails instead of halting the entire graph; they need to recover from failures without re-running jobs that already succeeded; and they need concurrency controls to stay within solver license limits.

These features bring Rescale Workflows in line with the reliability controls that production simulation teams expect from a workflow orchestration system.

How is it used? By whom?

• Simulation engineers running multi-branch workflows can let healthy branches finish even when one branch fails, and retry failed runs without starting over.
• Workflow template builders can configure per-node failure behavior and concurrency limits in the template so end users get the right defaults without manual setup.
• Teams with license constraints for commercial solvers (ANSYS, Abaqus, LS-DYNA, etc.) can cap concurrent nodes to stay within their license count.

Getting Started​

To configure node-level failure controls:

• Open a workflow template and select a node.
• Scroll to the Node Properties section in the configuration panel.
• Enable Continue workflow execution if this node fails.
• Repeat for each node, or open Workflow Settings (gear icon) and click Set For All Nodes to apply it to every node at once.

To set a concurrency limit:

• Open a workflow template and click the gear icon to open Workflow Settings.
• Enable the Concurrent Nodes toggle.
• Use the stepper to set the maximum number of nodes that can be active at the same time.

To retry a failed workflow run:

• Open a workflow run that has failed or partially failed nodes.
• Click Retry in the upper right corner.
• Review the color-coded node status to see what will re-run.
• Edit the template to fix the configuration that caused the failure.
• Click Retry Workflow to launch the continuation run.

For additional information, check out the Workflows documentation.

Release Highlights​

What is it?

Auto Tune is an automated hyperparameter optimization system for AI Physics surrogate models. When you create a new model, select Auto Tune as your configuration method, choose a search duration, and click Start Auto Tune. The system handles learning rates, batch sizes, architecture parameters, and optimizer settings automatically.

Auto Tune runs in three phases:

• Explore & Narrow Down — Quick training runs evaluate many configurations. Weak candidates are dropped early to save compute.
• Train Top Models Fully — Surviving candidates receive full-length training to reach their true potential.
• Evaluate Winner — Final accuracy test on held-out data selects the best model.

You monitor progress in real time on the Auto Tune Evolution tab, which shows models tried, models advanced, current best accuracy, GPU memory usage, and search progress. A Leaderboard ranks candidates by validation accuracy so you can see which models advanced, which were dropped, and which are still training.

Search duration options:

OptionDurationBest for****Quick Check1 hourVerifying your data pipeline works, getting a rough baselineBaseline Search (Recommended)6 hoursMost use cases — good balance of exploration and costThorough Search24 hoursLarge or complex datasets where marginal accuracy gains matterCustomSet your ownWhen you know your compute budget or have specific time constraints

Why did we build it?

• Eliminate manual tuning: No more creating one model at a time, tweaking parameters, and manually comparing results.
• Weeks of work in hours: Auto Tune explores configurations in parallel — work that would take weeks of manual experimentation — and completes a typical search in 6 hours.
• Cost-predictable: Search cost is bounded by time budget × hardware rate. For example, a 6-hour Baseline Search on Grossular-4 ($6.87/hr) costs at most ~$41.
• Learn from past runs: Compatible Prior Runs mode seeds new searches with hyperparameters from successful prior Auto Tune runs in your organization, so you benefit from your team’s history. The current run’s best hyperparameters are contributed back for future use.

How is it used? By whom?

• Simulation engineers select Auto Tune when creating a new model in AI Physics. Choose an architecture, pick a search duration, select GPU hardware, and click Start. Auto Tune runs autonomously — when complete, review the R² score in the Versions table to confirm quality before deploying for inference.
• Teams with prior training history can enable Compatible Prior Runs in Advanced Settings to seed hyperparameters from prior successful runs on similar datasets, narrowing the search space and converging faster.
• Power users can expand Technical Details during a search to view the Efficiency Frontier (accuracy vs. training cost) and the Promotion Ladder (rung-based advancement system) for deeper insight into the search process.

Monitor Auto Tune progress in real time — see models tried, models advanced, and current best accuracy on the Auto Tune Evolution tab.

Getting Started​

• Navigate to AI Physics > Models and click + New Model.
• Select Auto Tune as your configuration method.
• Choose your model architecture (DoMINO, TwoStreamTransformer, GeoTransolver, or MeshGraphNet).
• Set your search duration — Baseline Search (6 hours) is recommended for most use cases.
• Select GPU hardware and click Start Auto Tune.

When complete, check the R² score in the Versions table. A value close to 1.0 means the model is ready to deploy for inference. If R² is low or negative, try increasing search duration, adding more training data, or switching to a different architecture.

Release Highlights​

What is it?

Four curated workflow templates, maintained by Rescale, are now available to all platform users. When you click + New Workflow, a dedicated “Rescale Templates” section appears below your organization’s templates with a preview canvas so you can see each template’s structure before selecting it. In addition to running these Rescale templates as-is, you can also duplicate and customize them for your use cases.

The four templates are:

• Pre- and Post-Processing Pipeline — A fan-out/fan-in pattern where a pre-processing job generates multiple input files, parallel solver jobs process them simultaneously, and a post-processing job consolidates the results. Ideal for maximizing throughput on multi-variant simulation runs.
• Dependent Job Configuration — A Decision node evaluates upstream job metadata (via Custom Fields) and routes execution to the right hardware configuration based on problem size. Automatically right-sizes compute resources so you stop over-provisioning small jobs or under-provisioning large ones.
• Admin Template / Job Configuration Management — A chain of Decision nodes routes uploaded input files to pre-configured solvers (Abaqus, LS-DYNA, Python) based on file type. Template builders lock down approved software versions and hardware; end users just upload a file and click Run. Built for license compliance, cost control, and team onboarding.
• Agent-Enhanced Workflow — An Input Validator agent checks your file before the solver runs, and a Troubleshooter agent automatically diagnoses failures and proposes fixes. Catches invalid inputs before compute resources are allocated, saving time and cost.

Why did we build it?

Building a multi-node workflow from scratch requires understanding node types, connections, parameters, and execution patterns. Customers told us they wanted vetted starting points they could learn from and adapt — not just documentation, but working examples they could run immediately. These templates reduce the time from zero to a working workflow from hours to minutes.

How is it used? By whom?

• Simulation engineers can use templates as a starting point to run common pipeline patterns (parallel processing, conditional routing) without building from scratch.
• Workflow authors and template builders can duplicate templates and customize them — add solvers, change hardware, modify parameters, and share their versions with their team.
• IT admins can use templates to enforce standardized solver configurations and control which software versions and hardware are available to end users.

Browse and preview Rescale-provided workflow templates directly from the workflow creation flow.

Getting Started​

To run a Rescale workflow template

• In the left navigation, click Workflows.
• Click + New Workflow.
• Browse the Rescale Templates section in the template picker.
• Click Use Template to run it with your own inputs.

To review or modify a Rescale workflow template

• In the left navigation, click Workflows.
• Select a Rescale template from the table to review the template configuration.
• Click Duplicate to create an editable copy you can customize.

For additional information, check out the Workflow Templates from Rescale documentation.

Highlights​

• A balance you can actually see: Total AI Credits, consumed to date, remaining, and percent consumed, in one card per product at the top of the dashboard
• Day-by-day burn chart: See which days drove consumption, with a separate colored series per product, and switch months to compare a heavy week against a quiet one
• Consumption by user: A ranked table of who used the most credits in the selected month, broken out by product, so you can tell the difference between broad team adoption and one large exploratory session
• Three months of history: Prior-period totals sit next to the current month, giving you a trend instead of a snapshot
• Credit-level detail: Every credit on the account with its product, issue date, original amount, consumed and remaining balance, and expiration date — plus a per-credit CSV export for your own reporting

What You Can Answer Now​

The dashboards are built around common credit-related questions:

• “How many AI Credits does this workspace have left, in each pool?”
• “Are we on pace to run out before the license renews?”
• “Who on my team is using the Assistant and agents the most?”
• “Is AI Physics training or Assistant usage driving our burn?”
• “Was last month heavier or lighter than this one?”
• “When does this credit expire, and how much of it is left?”
• “Can I get the raw consumption numbers into my own spreadsheet?”

Get Started​

Open the Organization Administration Portal or your Workspace Administration Portal, then go to Dashboards → AI Credits Utilization. Use the month picker in the title bar to change the period. The Consumption by User table has an export link to pull the same data as a CSV.

For credit-level detail, go to Financials → AI Credits. Each row is one credit — a license allocation or a top-up — with its own CSV link for a per-credit consumption report.

Organization admins see credit data aggregated across all workspaces in the organization, with a Workspace column identifying where each user’s consumption came from. Workspace admins see their own workspace only. No setup is required — access follows your existing admin role.

For complete documentation, see the AI Credits documentation.

The AI Credits Utilization dashboard: balance card, daily burn chart, consumption by user, and prior periods

The AI Credits page under Financials, showing each credit with its remaining balance and expiration date

Good to Know​

Numbers update on a daily cycle, not instantly. Consumption is rolled up several times a day, so a conversation you just finished with the Assistant may not appear in the charts or the CSV until the next rollup. The balance figure accounts for the current day’s usage; the per-user tables and consumption reports reflect completed rollups. If you are checking whether a specific interaction was tracked, look the following day.

Credits are shared across the workspace. Neither pool is divided per person — any user in the workspace draws from the same balance, which is why the Consumption by User table matters. Per-user caps are not part of this release.

Watch your email as well as the dashboard. Admins receive notifications at 50%, 80%, and 100% consumption. When a pool reaches zero, the features it covers stop for everyone in the workspace until credits are added, so treat the 80% notice as your cue to talk to your Rescale account team about a top-up.

The two pools burn differently. Agentic Engineering credits are consumed variably by the size of each Assistant or agent interaction — a quick question costs a fraction of what a long analysis over a large body of context does. AI Physics credits are consumed at fixed hourly rates for training, so its line on the burn chart tracks hours run rather than conversation length. Expect the two series to look differently.

Each pool is spent separately. Agentic Engineering and AI Physics credits are tracked independently and cannot be spent interchangeably — running out of one pool does not affect the other pool. The dashboard shows both cards side by side and the tables break consumption out by product.

Feedback​

Questions about credit balances, top-ups, or anything that looks wrong on either page: contact your Rescale account team or support.

License Hosts have supported maintenance windows since August 2025. This release brings the same control to Company License Proxies.

Highlights​

• Pick your own weekly window: Choose the day of the week and the hour (UTC) when maintenance may begin on each License Proxy, set independently per proxy.
• Required for FedRAMP Moderate organizations: If your organization has one or more FedRAMP Moderate compliant workspaces, your License Proxies now have a maintenance window. Every other organization can configure one optionally, and we recommend it — a proxy without a window does not receive security updates.
• More reliable window changes: Moving a proxy's window to a different day or hour now saves correctly every time. If you changed a window before June 2026 and found it reverted afterwards, set it again and confirm it holds.

Getting Started​

Open your Organization Administration Portal and go to Integrations → License Proxy, then select the proxy you want to configure. In the Maintenance Window Schedule section, click Edit, choose the day of the week and the hour, and click Update.

Times are 24-hour format and always UTC. They do not follow your local time zone and do not shift with daylight saving time. Pick an hour when you do not normally have long-running licensed jobs in flight. If you have several proxies, consider staggering their windows.

The day and hour you choose is when maintenance may begin. Updates are applied only if any are pending, so a reboot does not happen every week, and Rescale does not send advance notice when one does. That is why the window you choose matters.

For full details, see the License Hosting and Maintenance Windows documentation.

The Maintenance Window Schedule section on a License Proxy or License Host

Introducing Transient FEA Support for Crash in AI Physics​

Release Highlights​

What is it? The AI Physics OS now supports transient FEA as a use case for surrogate model training and inference. Engineers can prepare crash simulation results as AI Datasets, train models using the GeoTransolver architecture, and run inference to predict structural response across the full simulation time sequence.

Why did we build it? Crash simulation is among the most computationally expensive analyses in structural engineering — a single high-fidelity run can take hours to days, and design exploration requires many variants. What makes crash uniquely challenging for surrogate modeling is that the model must predict results at every time step of the simulation, not just a steady-state endpoint. The GeoTransolver architecture is purpose-built for this: it handles the spatiotemporal complexity of transient simulations, making it well-suited for crash and other time-dependent FEA problems. Replacing or augmenting full-fidelity crash runs with a trained surrogate model dramatically reduces the compute cost of design exploration.

How is it used? Simulation engineers working on crash or other transient FEA use cases can follow the new end-to-end crash tutorial, which demonstrates how to prepare OpenRadioss simulation results, create an AI Dataset, and train a GeoTransolver-based surrogate model. The trained model supports rapid inference — enabling design exploration at a fraction of the cost of running full simulations.

Comparison of OpenRadioss crash deformation results between the simulation results and surrogate model predictions

Getting Started​

Contact your Rescale representative to enable AI Physics for your account.

Follow the step-by-step crash tutorial: OpenRadioss AI Physics Tutorial.

For general guidance, see the AI Physics Featured Guide.

What's New​

Collections group compatible coretypes so the platform can allocate the best available hardware while holding your price fixed. You always select and are billed for the Collection, but until now, you could not see which specific member coretype actually ran. That detail is now visible:

• Jobs list Hardware column: Shows Collection (Actual Coretype) for completed jobs, for example "Emerald Plus (Emerald Max)"
• Hardware Settings page: The Collection tile displays "Ran on: Emerald Max" (or whichever coretype ran) beneath the Collection name
• Setup > Review page: After completion, shows Collection → Actual Coretype
• Job CSV export: Two columns distinguish what you selected (hardware_setting) from what ran (coretype_name)

Why It Matters​

The actual coretype matters for performance analysis. If you are comparing runtime across jobs or debugging a solver's behavior, knowing whether the job ran on the entry coretype or a higher-spec member helps explain differences. The Collection remains your product identity and your billing entity; the actual coretype is operational detail you can now inspect.

How to Use​

No setup required. The attribution appears automatically on any Collection job submitted after this release. For jobs submitted on a single coretype rather than a Collection, hardware_setting and coretype_name carry the same value.

See Coretypes for full details on Coretype Collections and attribution.

Overview​

Rescale Savings Plan is now available — a flexible, dollar-based commitment that applies savings automatically across any eligible coretype and region. Commit to a fixed $/hour spend level, and the platform applies discounts to eligible workloads without per-coretype contracts or region restrictions.

Highlights​

Dollar-Based Commitment​

Instead of committing a core count to a specific coretype, Savings Plan lets you commit a dollar-per-hour spend level (e.g., $10/hr). Discount percentage varies by coretype and is visible in the admin allowlist configuration. The platform applies savings to eligible workloads up to that hourly commitment, across any eligible coretype. Key capabilities include:

• One commitment covers all eligible coretypes, no per-type contracts to manage. Effective savings vary by coretype.
• No need to forecast which coretype you’ll use, the commitment adapts to your workload mix

Configuring a job with a Savings Plan eligible coretype for flexible savings.

Multi-Region Coverage​

Savings Plan applies regardless of which region your workloads run in. There is no region lock and no regional availability toggle to configure. Key capabilities:

• Workloads in any supported region are eligible
• No need to predict or constrain where jobs will run
• Eliminates the region-specific planning required by per-coretype commitments

Multi-Coretype Eligibility​

Savings Plan covers a defined set of eligible coretypes configured at purchase time. As workloads run on any eligible coretype, the platform applies full negation to covered workloads, depleting your hourly commitment automatically. Key capabilities:

• Eligible coretype list is configured by your Rescale account team
• List can be updated without a new contract as your workload profile evolves
• Covers both existing and newly-added coretypes on the eligible list

Same Platform Experience​

Savings Plan shares the same admin and user experience as the Coretype Plan:

• Queue policy: Immediate, Balanced, and Savings Required modes available
• Utilization Dashboard: tracks $/hr consumed vs. committed over time
• Workspace scope: apply to all workspaces or restrict to specific teams
• Invisible orchestration: users select ODP; savings apply automatically
• Billing: same “Applied Commitment” negation in Billing Summary

Rescale Savings Plan Utilization Dashboard

Why This Matters​

• Simplifies commitment management: one plan covers your entire eligible workload instead of managing multiple per-coretype contracts.
• Reduces forecasting burden: commit a dollar amount, not a per-coretype core count. Your commitment adapts as your workload mix shifts.
• Covers multi-coretype and multi-region workflows: customers running diverse simulations get savings without pinning to a single hardware choice.

Get Started​

Interested in a Rescale Savings Plan? Contact us or reach out to your Rescale account team to discuss commitment sizing and eligible coretype configuration.

For additional information, see the Commitment Plans documentation.

Highlights​

• Multi-node MPI support: Submit jobs across multiple nodes on CoreWeave using the same workflow as any other Rescale hardware
• Kubernetes-based orchestration: Rescale uses Kubernetes for cluster management to provision and manage the environment needed for distributed jobs to scale across nodes without added infrastructure complexity.
• Lower infrastructure cost: CoreWeave offers meaningful cost savings for CPU workloads on reserved instances compared to traditional hyperscalers
• Production-ready: Tested end-to-end with enterprise customers’ CAE simulations running at scale, with multi-node configurations validated
• Broad simulation support: Optical simulation, CFD, and structural analysis workloads are all supported, including large parallel jobs that run for hours

Getting Started​

Contact your Rescale account team to enable access to CoreWeave in your environment. Once enabled, select CoreWeave hardware when configuring your job and submit normally. You do not need to change your job scripts or MPI configuration.

Multi-node is available today for CPU workloads, best for large, long-running parallel jobs.

What Is It​

A knowledge base indexes the content from your connected sources including PDFs, Word files, presentations, and other engineering formats, so the Rescale AI Assistant can read, understand, and answer questions about them. Instead of searching for files by name and opening them manually, you can ask the AI assistant a question and get an answer drawn directly from your organization’s documents, with citations back to the source.

With data connectors, users can already browse and attach files from connected sources. Knowledge base goes a step further: it lets Rescale Assistant search inside those documents and surface the information you need.

How Engineering Teams Use Knowledge Bases​

Ask questions, get grounded answers: Instead of manually searching through documents, ask the AI Assistant directly: “What are the recommended boundary conditions for external aerodynamics?” The Assistant retrieves relevant content from your indexed documents and generates an answer citing the specific source files.

• Semantic search, not just keywords: The knowledge base understands meaning. A search for “thermal analysis results” finds documents about “heat transfer simulation outputs” even if those exact words don’t appear.
• Citations you can trace: Every answer includes citations referencing the source document and location. You can trace any claim back to the original file.
• Control what gets indexed: For SharePoint connectors, admins can configure path restrictions to scope the knowledge base to specific document libraries or folders, useful for large sites where only a subset of content is relevant.

Rescale Assistant Summarizes Information from Connected Sources

Supported Sources for Knowledge Base​

• AWS S3 Object Storage
• Microsoft Azure Blob Object Storage
• Microsoft SharePoint Cloud Storage

Supported Document Types​

Common engineering file formats are supported, including PDF (with images and tables), Word documents, PowerPoint presentations, plain text, and XML (including PLMXML).

Getting Started​

To enable and begin using these new capabilities, follow these steps:

1. Admins: Connect a source (S3, Azure Blob, or SharePoint) via Connectors > Manage Connectors. Click “Sync” to trigger ingestion. For SharePoint, optionally configure path restrictions to scope indexed content.
2. Users: Open the AI Assistant and ask questions about your organization’s engineering documents.

Requirements​

• Workspace must have the Agentic Engineering platform subscription enabled. Please contact your account manager for more information.

Introducing the New Rescale Experience as Default​

What’s changing​

Your account will automatically open in the new interface. All of your existing data — saved filters, tags, custom fields, job history, and workstation configurations — carries over with no action required on your part.

What’s new in the interface​

• Workspace Folders: Organize jobs and workstations into nested project hierarchies. Share entire folders with granular permissions.
• Persistent sidebar navigation: Access Jobs, Workstations, Storage, Files, and Admin settings from any page in one click.
• Agentic Engineering: Deploy pre-built simulation agents for job troubleshooting, report generation, and workflow automation via the Agent Library.
• AI Physics: Train, validate, and deploy surrogate models from raw simulation data — accessible from the sidebar.
• Workflows & Data Connectors: Connect enterprise data sources and encode validated simulation processes into repeatable, multi-step workflows.

These capabilities are available exclusively in the new interface. Contact your account team to learn more about adding new product offerings.

The Classic UI remains accessible through August 31, 2026, via your user menu at the bottom of the sidebar. Hover over your email address and select “Classic UI” to switch back at any time.

Adapting to the new UI​

Tips for getting started​

• The left sidebar is collapsible — click the menu icon at the top to maximize your screen space.
• Click your workspace name at the top of the sidebar to access your folder structure.
• Some settings (User preferences, Admin portal access) have moved to the user avatar in the bottom-left of the UI
• Use the “Submit Feedback” button in the sidebar to report issues or suggestions.

Getting help​

• Read the full guide: New Rescale Experience Guide
• Contact support: support@rescale.com
• Reach out to your account team for questions about new product capabilities.

What You Can Do​

• Browse files from connected sources: Navigate the folder structure of any connected source directly from the Rescale platform. View file names, sizes, and folder hierarchy – the same structure you’d see in the source system.
• Search for files by name: Find files across a connected source using a file path search. Results show file name, path, and size.
• Attach connected files as workload inputs: For S3 and Azure Blob connectors, attach files directly from connected storage as inputs during job or workstation setup. No download and re-upload – files transfer to the compute cluster automatically at launch.

Searching & Browsing Files from a Connected Source

Supported Connectors​

See the table below for supported connectors and their corresponding features:

Additional connector types (PLM systems, requirements management tools) and feature expansion are in development for upcoming releases.

Getting Started​

1. Admins: Navigate to Connectors > Manage Connectors to add a connection. Provide credentials for your source system and test the connection.
2. Users: Navigate to Connectors > Browse Files to find and interact with files from connected sources.

Creating & Managing Connectors

Requirements​

• Workspace must have the Advanced Modeling & Simulation or Agentic Engineering tier enabled. Please contact your account manager for more information.
• Admin credentials for the storage you want to connect.

What is it and why did we build it?​

Simulation engineers and admins spend 30–40% of their time on repetitive expert tasks: triaging failed jobs, configuring workloads, organizing files, generating reports, and searching documentation. Agents automate these workflows so your team can focus on engineering decisions rather than platform operations.

Rescale Agentic Engineering provides a library of specialized agents available across two subscription tiers:

Advanced Modeling & Simulation Agents​

• Job Failure Summarizer: Reads failed job logs and returns a plain-language diagnosis, including root cause, file and line reference, and recommended next steps.
• Coretype Recommender: Returns a hardware recommendation for a given solver and workload based on Rescale’s platform-wide run history, pricing, and regional availability.
• Budget Control: Analyzes billing data through chat: answers budget questions, forecasts burn rate, explains spend changes, and recommends cross-team reallocations.
• Job Organizer: Inspects jobs in bulk, proposes a folder hierarchy and tag taxonomy along the strongest organizational axis, and applies it on your approval.

Agentic Engineering Agents​

• Knowledge Base: Indexes connected knowledge bases (solver manuals, input file specs, org references) and returns ranked, source-cited passages via semantic search. Serves as the shared retrieval layer for other agents.
• Job Troubleshooter: Closes the loop from failure to a running job: diagnoses the issue, proposes a fix, applies it with your approval, resubmits, and iterates until the job runs. Configurable as a Workflow Builder node.
• Input Validation: Reads an input file before submission, identifies the solver and analysis type, and returns a severity-ranked report with line numbers and suggested fixes.
• Report Generator: Populates a PowerPoint template with a campaign’s results, contour plots, findings, and recommendations, producing a first-draft engineering report in your org’s format.

Getting Started​

Users can get started with agents through multiple methods:

• The Agent Library via the left navigation sidebar
• Natural-language prompts in the Rescale Assistant
• Configured as nodes in the Workflow builder to automate parts or full end-to-end processes
• Context-based triggers throughout the platform, e.g. Troubleshoot failed jobs from a job’s ‘Results’ page

Agent availability depends on your subscription tier (Advanced Modeling & Simulation or Agentic Engineering). Contact your Rescale account team for details.

For additional information, check out the Rescale Agents documentation.