What You Can Do

A connected assistant works with the same objects the platform UI shows: sources, datasets, the ontology, Digital Twins, simulations, reports, and applications. Ask in plain language; tool names below are what the assistant reports while working, not something you type.

Terminology

The same two words mean the same thing everywhere — the UI, the REST API, and these tools. A Source is raw ingested data (an uploaded file, a connector import, an extracted document), identified by sourceId. A Dataset is a derived, queryable table over one or more sources (filters, joins, aggregations, SQL), identified by datasetId.

Sources belong to the organisation and are linked into workspaces; the same source can appear in several. Everything else lives in a single workspace. If a source is not yet linked into the workspace, the assistant links it first (link_source_to_workspace).

Explore what a workspace contains

• "What workspaces do we have?" (list_workspaces)

• "What is in this workspace?" for a one-call inventory of sources, datasets, ontology, and twins (get_workspace_overview)

• "What data exists in the org that is not linked here?" (list_org_sources)

• "Show me the schema of the shipments table" (get_source_schema)

Ask questions of your data

• "Revenue by region last quarter" becomes an Anchor SQL query over ontology concepts — the ontology plans the joins and time alignment (query_data)

• Raw SQL against the underlying sources, for the queries Anchor SQL cannot express (query_sql)

• "Show me the first rows" (get_source_preview)

• "Chart churn by tenure" (create_chart)

For metric questions, the assistant reuses the workspace's defined metrics by name rather than re-deriving them, and states the definition it used.

Bring new data in

Each route produces a new source, and imports run in the background (check_background_runs):

• Tables from connected databases and storage (import_from_connector)

• External APIs (register_connector, import_source)

• Structured data out of documents in the workspace (documents_to_source)

• Web pages or a research prompt into a table, on deployments with web research enabled (research_to_table)

Shape data for analysis

The server distinguishes three ways of combining sources: joins (widening on a shared key), concatenation (stacking same-shaped files), and time alignment (bucketing or as-of matching). The assistant:

• Probes real values before trusting a join key (probe_join_keys)

• Builds and modifies datasets (build_dataset, modify_dataset)

• Maintains the ontology that makes columns joinable across sources (query_ontology, modify_ontology)

Do causal work

• Run causal discovery on a dataset (run_causal_discovery)

• Create and train Digital Twins (create_digital_twin, train_digital_twin)

• Inject domain knowledge: forced edges, temporal order (update_digital_twin_relationships)

• Review a trained twin's quality before trusting it (review_digital_twin)

Discovery and training are background jobs: the assistant starts them and reports back when they finish.

Simulate and predict

• "What if we raise price 10 percent?" (query_digital_twin)

• "What drives churn?" (query_causal_graph)

• Interventions, optimisations, forecasts, root cause analyses

• "Does marketing drive demand in every region?" across a multi-environment twin (compare_environment_graphs)

Produce reports and applications

• Analytical reports from workspace evidence (create_report), revised in place (edit_report)

• Interactive Applications: runnable flows with dashboards, forms, and viewer-adjustable simulations (create_application, publish_application). See Applications.

Monitor usage

Ask usage questions directly: AI spend over time, top spenders by user or workspace, failure rates, security-relevant events (query_usage_monitor). Reads the same organisation audit log as the Usage Monitor page.

Pacing

• Search before import. The assistant checks existing sources, datasets, and registered connectors before fetching anything external.

• Background jobs return run ids, not results. The assistant polls for completion before claiming the data or model exists.

For which actions need your confirmation, see Permissions and Safety. For the packaged procedures behind the involved flows, see Guided Skills.