Demo workspace

Import the synthetic support-agent demo, verify each seeded asset, and understand where sample data stops and your own integration begins.

The demo import gives an empty organization a synthetic tracing-to-evaluation example. It creates a project named Demo so you can learn navigation before your own application sends telemetry.

Start the import

When creating an organization, leave Add sample data to explore selected.

For an existing organization with no traces and no demo project, use the Import demo data banner when it appears. Dismissing the banner hides the offer in that browser for 30 days; it does not create or delete data. The offer stops appearing after the organization receives real traces.

The import runs in the background. You can navigate while it completes, but counts and snapshot-backed cards can appear after the project row.

What the demo creates

Asset	Expected synthetic content
Project	Demo, project ID demo, environment production.
Traces	40 support conversations from 10 fixed scenarios across a 90-minute window.
Trace metadata	Service demo-support-agent, service version 1.4.0, users, sessions, model, retrieval and tool examples, tokens, latency, and estimated cost.
Dataset	Published Demo support conversations, slug demo-support-conversations, type agent.
Dataset items	36 items, one for each successful trace. Four intentionally errored traces are excluded by the mapper.
ETL	Configured Demo conversation mapper, code version 1, and a completed run over 40 traces.
Evaluation	Enabled Demo dataset evaluation, a completed run over 36 items, and 12 scorer configurations with a pass threshold of 7.
Scorer results	432 deterministic sample results: 36 items multiplied by 12 scorers.
NovaPilot	One completed recommendation report based on the demo dataset and results.
Analytics	A project analytics and Health snapshot refresh is queued after seeding.

The scenarios cover refunds, order status, password reset, shipping delay, cancellation, product recommendations, a billing dispute, API authentication failure, SDK troubleshooting, and a plan upgrade. Each appears four times. The four API authentication traces intentionally have error status, which is why 40 traces become 36 dataset items.

Verify the import

Open Projects > Demo after the progress indicator finishes.

1. Check project and trace scope

Project Overview should show activity. Open Traces, set the environment to production, and confirm that the project-scoped table contains the synthetic support traces. Four support.conversation.api_error traces should have error status.

Use Project Traces to inspect the flat table and detail sheet or Explorer to learn the organization directory.

2. Check the dataset and Runs

Open Datasets and select Demo support conversations. Confirm the published version and 36 items. Then open Runs, choose the same dataset, and compare a mapped transcript with its checks and source trace.

Runs reads the latest available dataset-scoped results. It does not create new evaluation results. See Review evaluated Runs.

3. Check ETL

Open ETL Jobs > Demo conversation mapper. The completed run should report 40 processed traces and 36 created items. The difference is expected because the mapper excludes the four intentionally errored conversations.

Open the mapper editor and preview a synthetic trace before changing code. Preview runs the editor content against the selected trace; saving creates or updates the mapper version. See Build and test an AI mapper.

4. Check evaluation

Open Eval Jobs > Demo dataset evaluation, then open its completed run and Results. The sample uses deterministic seeded results rather than live judge-model calls during import.

There should be 432 scorer results when all 36 items and 12 configured scorers are present. That number is scorer-item outcomes, not eval-job executions. See Set up, run, and review an eval job.

5. Check NovaPilot

Open NovaPilot > Reports and select the completed demo report. It summarizes seeded results and offers recommendations for investigation.

NovaPilot does not edit code, change prompts, or open pull requests. Your team decides whether to apply a recommendation through its normal ticket, branch, review, and deployment process. See NovaPilot recommendations.

6. Check snapshot-backed views

Project Health and organization Analytics can finish after the other assets. Use Update now when the current snapshot predates the import, then wait for the update before comparing counts.

Route map

Replace {organizationSlug} with the slug in your application URL.

Destination	Route
Demo project	/app/{organizationSlug}/projects/demo
Project Traces	/app/{organizationSlug}/projects/demo/traces
Demo Runs	/app/{organizationSlug}/projects/demo/runs/demo-support-conversations
Demo dataset	/app/{organizationSlug}/projects/demo/datasets?dataset=demo-support-conversations
ETL jobs	/app/{organizationSlug}/projects/demo/etl-jobs
Eval jobs	/app/{organizationSlug}/projects/demo/eval-jobs
Project Health	/app/{organizationSlug}/projects/demo/health
NovaPilot reports	/app/{organizationSlug}/projects/demo/novapilot/reports
Organization Analytics	/app/{organizationSlug}/analytics

ETL job, eval job, and report detail routes contain generated IDs. Open those records from their list pages rather than constructing a URL.

What is not created

The demo does not create:

• an SDK connection to your application or proof that your API key works;
• provider credentials, organization API keys, members, or invitations;
• custom scorers, runtime policies, alerts, or schedules;
• NovaSynth connections, personas, scenarios, test matrices, or runs;
• production data or live model calls during import;
• a durable evaluation policy for your product.

If the import looks incomplete

1. Wait for the background progress indicator to finish.
2. Refresh the Projects directory and the relevant list page.
3. Check the project, traces, dataset, ETL run, eval run, report, and snapshot in that order.
4. Do not repeatedly start the import or create a second project named Demo.
5. Contact support@noveum.ai with the organization slug if assets remain incomplete.

When you are ready for real data, send one trace, verify it in Project Traces, and follow The closed-loop workflow.