Demo Workspace

SOIT ships deterministic seed scripts that populate a demo workspace with a working agent, knowledge base, workflow, and runtime history. Use them to explore the product's core loop — model configuration, knowledge retrieval, agent execution, workflow execution, governed tool use, trace evidence, and observe views — without building everything by hand.

Seed the base demo data

After the stack is up and migrations have run, seed the demo data from the server/ directory of the SOIT repository:

cd server
uv run python scripts/bootstrap_enterprise_mvp.py

The seed is idempotent — running it again updates the same objects. It creates:

• A sample model provider with test models registered in ModelHub.
• A sample knowledge base containing a refund policy document.
• A sample agent bound to the model, knowledge base, and a governed tool.
• A sample workflow for support-ticket triage.

Seed richer runtime scenarios

For fuller Observe, Task, Run, approval, and failure-state demos, layer on the scenario seed:

uv run python scripts/seed_enterprise_mvp_scenarios.py --reset

This adds historical runs in different states (succeeded, failed, awaiting approval) so the Observe views have real data to show.

Verify the demo path

Run the backend smoke test that exercises the seeded agent, knowledge, and workflow path end to end:

uv run pytest tests/integration/test_enterprise_agent_mvp.py -q

You can also run the support-ticket regression evaluator, which produces a machine-readable report with citation, tool-call, child-workflow, audit, and cost evidence:

uv run python scripts/evaluate_support_ticket_regression.py --json-output ../artifacts/support-ticket-regression/report-current.json

The 20-minute governance tour

The demo workspace is designed to demonstrate the governance story quickly. A suggested route through the web UI:

1. Permissions — open /agents and inspect workspace-scoped agent bindings under the Owner role.
2. Secrets — open /settings/secrets to see the managed secret inventory used by governed tools and connectors.
3. Call audit — open /observe/audits?gateway_type=tool for policy-gateway request and response evidence.
4. Cost attribution — inspect run-level cost entries on any completed run.
5. Replay — open /observe/runs/{run_id} to see steps, tool calls, citations, audits, child runs, and costs for a single run.
6. Regression — the deterministic regression evaluator doubles as the agent publish gate.

Manual checks

• ModelHub shows the seeded provider and models.
• Knowledge shows the refund policy document.
• The sample agent answers a refund-policy question with citation evidence.
• The workflow executes the support-ticket triage path.
• Observe and run detail pages show response events, run steps, tool calls, child workflow runs, costs, citations, and audit records.

Next steps

• Understand what you just saw in Core: Build, Execute, Observe, Govern.
• Dig into individual product areas in Capabilities.
• Move to Development Setup if you want to change code.