Rescale Agents Product Documentation

Overview

Rescale Agents are purpose-built AI agents for simulation and HPC workflows. Each agent handles a specific task in the simulation lifecycle, from validating an input file before submission to diagnosing a failed job, benchmarking hardware, or generating a campaign report. Agents are available through the Rescale Assistant and the Workflow Builder, and they compose with one another through the Orchestrator so a single request can draw on several agents at once.

Agents are organized under two different product offerings: Advanced Modeling and Simulation covers the agents that help you run, recover, and analyze simulation work, and Agentic Engineering covers the agents that automate and govern the platform itself. Access to each is controlled at the workspace level by your administrator.

Highlights

  • Task-specific agents across the simulation lifecycle – validation, troubleshooting, benchmarking, reporting, organization, and more.
  • Composable through the Orchestrator – agents call one another, so complex requests resolve in a single conversation.
  • Available where you work – invoke agents from the Rescale Assistant chat or wire them into the Workflow Builder as nodes.
  • Grounded in Rescale data – recommendations and analyses draw on your organization’s run history and ingested knowledge bases, not generic defaults.
image 289

Using Rescale Agents

To invoke an agent, type `/` in the Rescale Assistant. A list of the available agents appears; select the one you want and pass it parameters or a prompt describing the task. You can also add an agent as a node in the Workflow Builder. The Orchestrator routes your request to the right agent (or chain of agents) and returns the result. The sections below list the available agents by feature set, with what each one does and the use case it addresses.

image 289

Prerequisites

Before you begin:

  • Your workspace must have the relevant feature set enabled: Advanced Modeling and Simulation or Agentic Engineering. Access is controlled by your workspace administrator.
  • You need permission to view the jobs, workspaces, or budgets an agent acts on.
  • Some agents (for example, Input Validation and Job Troubleshooting) expand their coverage through authored skills. Coverage for a given solver depends on what your organization has configured.

Advanced Modeling and Simulation Agents

These agents help you run, recover, and analyze simulation work and will be ever evolving as agents improve and as new agents are introduced. 

Job Failure Summarizing Agent

Reads a failed job’s solver logs and error output and produces a plain-language diagnosis: what the error is, which file and line surfaced it, and what it likely means.

Use case: Skip the 50,000-line transcript. An Ansys Fluent job fails overnight and the engineer gets a one-paragraph root cause instead of scrolling raw logs. Works through LLM reasoning over logs, so it returns a diagnosis even for unfamiliar solvers and failure modes.

Coretype Recommendation Agent

Returns an initial hardware recommendation for a given solver and workload, drawing on Rescale’s platform-wide run history.

Use case: Stop guessing at hardware. An engineer about to submit a large Abaqus job gets a recommendation in seconds instead of asking a senior colleague. Used standalone before submission, or as a sub-agent inside the Compute Configuration Benchmarking Agent.

Metadata Extraction Agent

Extracts metadata from existing jobs post-hoc, routing structured text through parsers for solver logs, CSVs, and config files, and grounds every value in a source file and line.

Use case: Make legacy jobs findable. An engineer wants “jobs with mesh > 5M cells using Fluent 2024R1” and runs the agent across a year of legacy jobs to back-fill the metadata, then filters in the UI. Fills the gap for jobs that ran before metadata automation was configured.

Job Organizer Agent

Inspects jobs in bulk, places each in the right folder along the strongest organizational axis (program, discipline, campaign), and tags cross-cutting dimensions.

Use case: Clean up a sprawling workspace at scale. A team inheriting an unorganized workspace gets it structured in an afternoon instead of weeks of manual triage. Tags only dimensions Rescale does not already capture as structured metadata.

Budget Control Agent

Analyzes Flexible Budgets and billing data through chat: answers budget questions, forecasts burn rate, explains why spend changed, recommends cross-team reallocations with dollar amounts, and creates or updates budgets via natural language.

Use case: Catch budget imbalances before the quarter closes. At the end of Q2 a manager asks “how are my teams tracking?” and gets a forecast plus a recommended reallocation. A finance partner asks why simulation spend jumped 40% and the agent attributes it to a shift to GPU jobs on a specific project.

Agentic Engineering Agents

These agents automate and govern the platform itself, and provide the composable infrastructure other agents build on.

Job Troubleshooting Agent

Closes the loop from failure to a running job: diagnoses the failure (using the Job Failure Summarizing Agent), proposes a fix, applies it with your approval, resubmits, and iterates until the job runs.

Use case: Automatic recovery on long campaigns. A crash simulation fails at 2am; the agent diagnoses a mesh quality issue, proposes a fix, asks the on-call engineer for approval, and has the job running before standup. Configurable as a Workflow Builder node that auto-triggers on failure.

Input Validation Agent

Reads your input file before submission, identifies the solver and analysis type, and returns a report with severity levels, line numbers, plain-language explanations, and suggested fixes.

Use case: Catch errors before they cost compute time. An engineer submitting a large LS-DYNA crash job catches a malformed contact card before burning 12 hours of solve time. Available as a button during job submission or as a Workflow Builder node.

Report Generator Agent

Populates a PowerPoint template with a campaign’s pass/fail matrix, contour plots, findings, and recommendations, producing a first-draft engineering report.

Use case: Turn results into a deck in minutes. A thermal campaign concludes Friday afternoon; Monday morning the engineer has a draft report to refine instead of a stack of raw results. Organizations can standardize on their own report templates.

Queue Intelligence Agent

Fetches job data, computes queue statistics (average, median, p95 wait times), parses scheduler logs for blocking reasons, and generates filterable trend charts.

Use case: Turn queue data into answers. A simulation admin asks why Fluent jobs are slow this week and gets “license-limited 80% of the time, here’s the wait distribution” – useful evidence for license renewal conversations. Chat-based, with no dashboard integration required.

Job-as-Code Agent

Turns a natural-language job description (or an existing job used as a template) into runnable API code – cURL, Python, or a requested language – that creates and submits a Rescale job end-to-end, resolving software and hardware codes against the live catalog.

Use case: Drop Rescale into your own scripts and pipelines. An engineer wiring Rescale into CI asks for a Python snippet that submits their nightly validation job and gets working code in one turn instead of reading API docs. Uses environment-variable placeholders, so the code is safe to commit.

Knowledge Base Agent

Indexes connected knowledge bases (solver manuals, input file specs, organization references) and exposes them via semantic search, returning the most relevant passages for a query.

Use case: The shared retrieval layer for other agents. A Skill Authoring Agent asks for the official Fluent input file spec before generating a validation skill; the Report Generator retrieves prior-iteration results for a campaign comparison. Composed by other agents through the Orchestrator whenever grounded, source-cited context is needed.

FAQ

How do I invoke an agent?

Type `/` in the Rescale Assistant to bring up the list of available agents, select one, and pass it parameters or a prompt. You can also add an agent as a node in the Workflow Builder. The Orchestrator routes your request to the right agent.

Can agents work together?

Yes. Agents compose through the Orchestrator. For example, the Job Troubleshooting Agent calls the Job Failure Summarizing Agent to diagnose before proposing a fix, and the Compute Configuration Benchmarking Agent calls the Coretype Recommendation Agent to narrow its sweep.

Why don’t I see a particular agent in my workspace?

Agent availability depends on which feature set your workspace has enabled (Advanced Modeling and Simulation or Agentic Engineering) and your permissions. Contact your workspace administrator.

What happens when an agent encounters a solver it doesn’t recognize?

Agents like Input Validation and Job Troubleshooting expand coverage through authored skills. When no skill exists for a solver, the agent directs you to author one, after which every engineer in your organization benefits.

Do agents change my jobs or budgets without asking?

Agents that take actions (for example, Job Troubleshooting applying a fix, or Budget Control modifying a budget) ask for your approval before any state change.

Relevant Links

×