Decision intelligence platform
Live metrics, LLM answers and ML forecasts on one decision board.
OrionPulse sits on top of the warehouse your data team already runs. Finance, retail and manufacturing teams ask questions in plain English, get forecasts they can check, and track every decision to a measured result.
14-day trial. Read-only warehouse access. No card required.
- Boards
- Ask Orion
- Forecasts
- Decisions
- Metrics
- Connections
Retail ops · weekly
Sample data · synced 4 min ago
- A · Net revenue (7d)
- $2.84M
- +3.1% vs prior
- Forecast accuracy
- 94.2%
- MAPE 5.8%
- Stockout risk SKUs
- 37
- 12 in Northeast
- D · Open decisions
- 9
- 2 due today
What each part of the board is doing
- A
Metric tiles read from the semantic layer
Net revenue, forecast accuracy and stockout risk are defined once in the semantic layer, so the tile, the LLM answer and the API return the same number.
- B
Forecast line with confidence bands
Forecast Studio trains gradient-boosted and time-series models on your history and redraws the line after every warehouse sync.
- C
Ask Orion, the LLM analyst
Type a question in plain English. Orion writes the SQL against governed metrics, shows the query, and runs it inside your warehouse under your permissions.
- D
Decision queue
When a metric crosses a threshold, a decision card opens with an owner, a due date and the evidence. The outcome is written back so you can measure what changed.
C · Question: Which Northeast stores will run out of top-50 SKUs before the next delivery?
SELECT store_id, sku, days_of_cover FROM metrics.inventory_cover WHERE region = 'Northeast' AND sku_rank <= 50 AND days_of_cover < next_delivery_days ORDER BY days_of_cover;
12 rows · ran in Snowflake in 1.4 s · metric: inventory_cover (verified)
D · Decision queue
| Decision | Owner | Status |
|---|---|---|
| Expedite replenishment 12 SKUs · Northeast | Supply planning | Open |
| Pause promo PRM-2210 Margin below floor | Merchandising | In review |
| Rebalance DC-04 stock Overstock 18 days | Logistics | Approved |
Data security, written as practices
Banks, retailers and manufacturers review us before connecting a single table. These are the controls they check.
Request the security questionnaire- Queries run in your warehouse
- OrionPulse pushes SQL down to Snowflake, Databricks or BigQuery. Raw tables are not copied into our storage.
- Encryption
- TLS 1.2+ in transit, AES-256 at rest for metadata, cached results and audit logs.
- Identity and access
- SAML 2.0 / OIDC single sign-on, SCIM provisioning, role-based access and row-level rules inherited from your warehouse.
- LLM data handling
- Prompts go to model providers under zero-retention API terms. Your data is never used to train any model.
- Audit trail
- Every question, generated SQL, export and decision change is logged with user, time and source. Logs export to your SIEM.
- Export and deletion
- Export boards, metrics and decision history as JSON or CSV at any time. Workspace data is deleted within 30 days of contract end.
From warehouse to first tracked decision
- 01
Connect your warehouse
Add a read-only service account for Snowflake, Databricks, BigQuery, Redshift or PostgreSQL. Schema discovery takes minutes.
- 02
Publish a semantic model
Define metrics, joins and business names once. Start from auto-generated drafts and review them with your data team.
- 03
Open Ask Orion and Boards
Business users ask questions and pin answers to low-code boards. No new SQL skills needed.
- 04
Turn thresholds into decisions
Set Decision Flows that open tasks, notify Slack or Microsoft Teams, and record the outcome.
Questions data leaders ask us
Do you copy our data out of the warehouse?
No. Queries are pushed down to your warehouse and only result sets needed for a board or answer are cached, encrypted, for up to 24 hours. You can turn caching off per workspace.
Which large language models does Ask Orion use?
Ask Orion routes to current models from Anthropic, OpenAI and Google, listed on the Features page. Enterprise customers can bring their own API key or a private endpoint on Amazon Bedrock or Azure.
How accurate is text-to-SQL?
Orion only generates SQL against metrics and joins in your semantic model, shows every query before it runs, and lets analysts mark answers as verified so they are reused.
How long does a typical rollout take?
Most teams connect a warehouse and publish a first semantic model in the first week of the 14-day trial. A guided onboarding is included on Business and Enterprise.
How do you measure ROI?
Each Decision Flow records the baseline metric, the action taken and the metric afterward. The ROI report sums the measured change per flow, so the value is tied to decisions you can audit.
Is there an API?
Yes. A REST API, webhooks, an Embed SDK and an MCP server are available. See the Developers page.