PREVIEW · static UI mock · screen 13 / 32 · workflow integrations · all screens
‹ Workflows

Demand prediction

● Active v23

Owner: Maria Chen · Supply Chain VP · 6 integrations connected · open agent UI →

Connected
6 / 6
3 sources · 3 outputs
Sync health · 24h
100%
no failed syncs
Data ingested · 7d
2.4M rows
SAP sales · NOAA weather · Snowflake
Output events · 7d
412
alerts + tickets + digests
Data sources 3 connected · what the agent reads from
SAP ERP · Sales history
Historical weekly sales by SKU-store · 4 years backfill
● Connected
Tool exposed query_sap_sales
Last sync 2h ago
Sync rate every 6h
NOAA Weather API
Regional weather forecasts + anomaly indices
● Connected
Tool exposed fetch_noaa_weather
Last sync 30m ago
Sync rate every 1h
Snowflake · SKU master
SKU catalog, store metadata, supplier lead times
● Connected
Tools exposed list_skus, get_inventory_status
Last sync 1d ago
Sync rate daily
Add a data source
Postgres, BigQuery, REST API, S3, Salesforce, Workday…
Outputs 3 active · where the agent writes / notifies
Slack · #supply-chain-alerts
Markdown alerts above medium severity
● Active
Tool exposed post_slack_alert
Fires · 7d 38
Last fire 14m ago
Email · daily digest
9am PT digest to Maria Chen + 4 stakeholders
● Active
Tool exposed send_digest
Fires · 7d 7
Last fire today 9:00 PT
ServiceNow · ticket queue
Open ticket on each high-severity markdown alert
● Active
Tool exposed open_servicenow_ticket
Fires · 7d 14
Last fire 2h ago
Add an output
Webhook, Teams, PagerDuty, Jira, custom API…
Each integration becomes a tool the agent can call (data sources are read tools; outputs are write tools). Tool calls are recorded in the trace and counted toward eval / regression scoring. Outputs are simulated in the sim environment — the agent can't actually post to Slack during eval runs.