About NVIDIA NeMo Studio | NVIDIA NeMo Helix

Studio is still in early development. Many features are missing or should be expected to change.

NeMo Studio is the web app for AI development with NVIDIA NeMo Helix. It provides a workspace-oriented UI for managing local platform resources, reviewing agents, running agent optimization workflows, monitoring agent telemetry, and working with filesets, jobs, and secrets.

The pages visible in your workspace depend on the installed plugins, enabled services, feature flags, and NeMo Helix version. A page documented here might be hidden in your deployment. Preview pages identify the feature flag that enables them.

Getting Started

NeMo Studio is included with the platform. Follow the Setup guide to start NeMo Helix, then access Studio at /studio on your running server.

Features

The current sidebar groups related pages under expandable parents. Selecting a parent that has its own landing page, such as Agents, opens that page; when it has child entries, use its chevron to expand them. A parent without a landing page, such as Models, only expands its child entries.

Sidebar group	Current entry point	Purpose and availability
Observability	Insights	Review optimization insights generated from observed agent sessions. Shown when optimization is enabled; see NeMo Studio Insights.
Observability	Traces	Inspect ingested traces, spans, and session details; see NeMo Studio Traces.
Components	Agents	Opens the agent list. Select an agent to access the Overview, Deployments, Logs, Chat, Evaluations, Optimizations, Insights, and Details tabs. Optimizations and Insights are enabled by default and can be hidden by their feature flags.
Components	Models	Expands to Model Catalog, Evaluations, Fine-tuning, Playground, Deployments, and Virtual Models, when those capabilities are enabled. Model Catalog opens the base-model list. Evaluations opens the Model Evaluations page; see NeMo Studio Model Evaluations. Fine-tuning opens the fine-tuned models list; see NeMo Studio Fine-tuning. Playground is disabled by default. Virtual Models appears only when the Guardrails preview feature is enabled.
Evaluations	Experiments	Group evaluations for comparison, drill into results, and compare runs side by side; see NeMo Studio Experiments.
Data	Datasets > Anonymizer	Detect and protect PII in a dataset. Enabled by default; see NeMo Studio Anonymizer.
Data	Datasets > Data Designer	Build and monitor synthetic-data jobs. Other installed data plugins can appear below Datasets.
Data	Datasets > Safe Synthesizer	Create and monitor privacy-preserving synthetic tabular data jobs. Enabled by default.
Governance	Guardrails	Manage guardrail configurations. Disabled by default; see Studio Guardrail Configs.
Governance	Agent Hardening	Plugin-provided agent hardening UI. Appears only when the Agent hardening plugin is installed; see Studio Plugin UIs.
System	Filesets	Organize and manage files used by agents and jobs.
System	Jobs	Monitor jobs across enabled platform capabilities.
System	Settings	View workspace settings and open Secrets and inference provider management. Enabled by default.

Dashboard appears above the sidebar groups and is shown by default. Playground and Virtual Models are disabled by default and appear only when an administrator enables their feature flags. Because plugins contribute navigation at runtime, an installed plugin can also add pages to an existing group or add a new group.

Agents

Use the Studio Agents area to review NeMo Helix-managed agents, including the current Fabric-first nemo-agents-spec-v1 experience, inspect agent details, deploy agents, open a chat session against a running deployment, and clean up deployments that are no longer needed.

For the full workflow, see Studio Agents.

Insights

Use Observability > Insights to review optimization insights the analyst agent generates from observed agent sessions and traces, and drill into an insight’s linked experiments and observed sessions. Use Resolve or Delete to close an open insight, and Reopen to restore a resolved or deleted one.

For the full workflow, see Studio Insights.

Traces

Use Observability > Traces to inspect ingested telemetry in the Traces and Spans tables, and drill into a session’s trace and span detail, with specialized views for LLM, tool, agent, chain, embedding, reranker, retriever, guardrail, and evaluator spans.

For the full workflow, see Studio Traces.

Guardrails

When the preview feature is enabled, use Governance > Guardrails to create and manage NeMo Guardrails configurations for the current workspace. Select Create Guardrail to add a configuration. The list shows each config’s Name, Main Model, Flows, and Updated columns, with an optional Created column; use the row actions to Duplicate or Delete a config, or select several rows to delete them together.

Select a row to open the detail page, which has Configuration, Test and Validate, and Virtual Models tabs. The Configuration tab lets you edit the main model and rails, and shows the full payload as read-only Configuration JSON. Select Save Guardrail to apply your changes or Reset to discard them. The detail page also offers Duplicate and Delete actions.

For the full workflow, see Studio Guardrail Configs.

Virtual Models

When the Guardrails preview feature is enabled, use Models > Virtual Models to inspect inference routes managed by the Inference Gateway. The list shows all user-created virtual models in the workspace — auto-provisioned passthrough routes are hidden. Select a row, or choose View from its row actions, to open the virtual model’s detail page. The Details tab shows the default backend model when configured, and the ordered request, response, and post-response middleware pipelines attached to the route. The Chat tab sends test prompts through the virtual model. Row actions also include Chat and Delete.

Virtual models are created through the nemo inference virtual-models CLI or the Inference Gateway API.

For the full workflow, see Studio Virtual Models.

Models

Select Models > Model Catalog to browse base-model entities available in the current workspace. Models is an expandable parent without its own landing page; Model Catalog is the base-model list. Base models include models registered in the workspace and default platform models.

The Model Catalog page supports searching by model name, filtering by creation or update date, sorting alphabetically or by creation time, and showing only customizable models. Select a model card to open the model details side panel.

The model details side panel has Model Details and Chat Playground tabs. Model Details shows model metadata, parameters, and available actions, such as Customize this Model when fine-tuning is enabled. Chat Playground lets you chat with chat-capable models.

Fine-tuning

Select Models > Fine-tuning to open the fine-tuned models list, start a fine-tuning job against an eligible base model, and track job progress. Enabled by default; see NeMo Studio Fine-tuning.

Workspaces

Keep models and data organized in dedicated workspaces.

Filesets

Use Filesets to organize files by purpose, which determines which metadata fields are available. Purpose is chosen at creation time and can’t be changed afterward.

Purpose	Use for
Generic	Files that don’t fit the Dataset, Model, or Environment categories. Doesn’t add purpose-specific metadata fields.
Dataset	Default for new filesets. Training and evaluation data. Enables dataset-specific metadata, including schema information. Use built-in tools to transform data schemas with LLM assistance or create train/test splits.
Model	Model weights and checkpoints. Enables model-specific metadata, including tool-calling and model configuration fields.
Environment	Executable environment packages consumed by sandboxed workloads.

Supported file formats: JSON (.json), JSONL (.jsonl), CSV (.csv), and Parquet (.parquet). PDF, image, and other binary formats are not currently supported.

You can upload any file type into a fileset. However, each service supports specific input file types:

Service	Supported File Types
Data Designer	`.json`, `.jsonl`, `.csv`, `.parquet`

Anonymizer

Use Datasets > Anonymizer to detect and protect PII in a dataset, choosing a substitute, redact, annotate, hash, or rewrite strategy, previewing the result on a few records, and running a full job. Enabled by default.

For the full workflow, see NeMo Studio Anonymizer.

Safe Synthesizer

Use Datasets > Safe Synthesizer to create and monitor privacy-preserving synthetic tabular data jobs from NeMo Studio. The Studio page provides access to job creation, job status, details, reports, and result downloads for Safe Synthesizer workflows.

For the full workflow, see Synthesize Safe Data.

Evaluations

Studio provides separate agent, model, and workspace evaluation views:

Location	What it shows
Agents > select an agent > Evaluations	In-progress and completed evaluations in one table, and experiments scoped to the selected agent. Start a new run with Run Evaluation.
Models > Evaluations	The Model Evaluations page: start a model evaluation with Run Evaluation, and review job results and per-row scores. See Studio Model Evaluations.
Evaluations > Experiments	Experiments and their evaluations across the workspace, evaluation drill-down, and side-by-side run comparison. See Studio Experiments.

Use the agent’s Evaluations tab when reviewing how a specific agent performed. Use Experiments for the broader workspace history.

Experiments

Use Evaluations > Experiments to run an evaluation against an agent from the UI — selecting the agent and judge model, uploading a dataset and evaluation config — and to compare results across runs on the experiment leaderboard.

For the full workflow, see NeMo Studio Experiments. For concepts, the data model, and the CLI and SDK workflows, see Experiments.

Jobs

Track and manage all platform jobs from a single view. The Jobs page aggregates jobs across enabled capabilities, such as Data Designer, so you can monitor progress, view logs, download artifacts, and cancel jobs in one place.

Navigate to Jobs in the workspace sidebar to see all jobs in the current workspace.

Search and filter — Find jobs by name, filter by status, or narrow results by creation or update date range.

Status tracking — Each job displays its current status:

Status	Description
Created	Job has been created but not yet scheduled
Pending	Job is queued and waiting for resources
Active	Job is currently running
Completed	Job finished successfully
Error	Job encountered an error
Cancelled	Job was cancelled by a user
Cancelling	Cancellation is in progress
Paused	Job execution is paused
Pausing	Job is transitioning to paused
Resuming	Paused job is restarting

Service routing — Clicking a job navigates to its service-specific detail view when one exists. Jobs from services that aren’t enabled fall back to the generic job detail view.

Cancelling a job — Active or pending jobs can be cancelled from the job list or the detail page. Cancelling a job permanently stops it and cannot be undone.

Secrets

Store API keys and credentials to securely connect with external providers. Open Settings in the System group and select Manage Secrets. See manage-secrets for details.