Orion's chart suggestions

Orion does not just answer in words. After it pulls data, it decides which chart type best fits the shape of the result and shows you a visual version of the answer in the same chat bubble. It also tells you how confident it is in that choice, and gives you one-click ways to keep the result — save it as a reveal or pin it to a mosaic.

This page is about the chart Orion suggests on its own for a fresh result. If you want to change a chart that already exists on a tile or in Discover, that is a different tool — see Orion chart copilot.

Overview

Every time Orion runs a run_reveal or run_sql query and gets rows back, it does two things automatically before it replies:

1. Picks a chart — it inspects the columns (metrics, dimensions, dates, geography) and the number of rows, then chooses a chart type such as a KPI, line, bar, map, or pie.
2. Scores its confidence — it attaches a confidence value (roughly 0 to 1) for that choice, so you know how strong the recommendation is.

Inside the chat, Orion renders the answer as a structured analysis block when it produces one: an overall banner (the headline number and its change), dimension bars (a tabbed breakdown showing each segment's contribution), and a short list of insights. That structured block is the in-chat "chart". Alongside it, any answer that ran a query carries a save bar so you can turn the result into something permanent.

When to use it

• You asked Orion a question and want to see the answer, not just read it.
• You want a quick read on whether a result is best shown as a single number, a trend line, a ranked bar list, or a map — without setting up a chart yourself.
• You like an Orion answer and want to keep it: save it as a reveal in your semantic model, or pin it to a mosaic.
• You want a fast confidence signal on whether the chosen chart actually fits the data.

Concepts

Getting started

• Access to Orion (Tracer, Weaver, or Luminary).
• A connected warehouse and at least one modeled domain so Orion has data to query.
• For raw-SQL answers, note that run_sql is governed by role — non-admin roles may be refused for certain raw queries, and Orion will say so honestly rather than return a blank answer.

Step-by-step

Examples

Ask Orion:

Show me revenue by region last 30 days

Because the result has one metric, one dimension, and more than eight rows, Orion suggests a horizontal_bar chart. With one metric and a dimension but eight rows or fewer, it would pick a regular bar instead. The answer in chat renders as a structured analysis — an overall banner for total revenue and a "By Region" bar breakdown.

Ask a trend question:

What's the daily revenue trend over the last 90 days?

Here the result has a metric and a date column with no breakdown dimension, so Orion suggests a line chart. Trend wording like "over time" or "last N days" also nudges Orion toward a line.

A single-number question:

What was total revenue yesterday?

One metric, one row, no dimension or date — Orion suggests a kpi and scores it 1.0.

When Orion writes a raw query, it returns it in a fenced sql block, and an Add to mosaic button appears under that block:

SELECT region, SUM(net_revenue) AS revenue
FROM orders
WHERE order_date >= CURRENT_DATE - INTERVAL '30 days'
GROUP BY region
ORDER BY revenue DESC;

Best practices

• Phrase the question to match the shape you want. Words like "trend", "over time", or "last 30 days" steer Orion toward a line; "share", "breakdown", "split", or "distribution" with a small result steer it toward a pie.
• Use Save as reveal for things you will ask again. A reveal lives in your semantic model and is governed, so it stays reusable and consistent.
• Use Add to mosaic for monitoring. Pin the result as a tile so it refreshes on the dashboard instead of re-asking Orion each time.
• Name the insight before saving. The save bar's name field becomes the reveal name and the tile title — a clear name saves cleanup later.
• Treat low confidence as a prompt to refine. If Orion falls back to a table (confidence 0.50), the shape was ambiguous; tighten the question or pick the chart yourself after saving.

Tips

Common mistakes

Troubleshooting

Related pages

• Orion chart copilot — edit an existing chart by describing the change.
• Asking Orion questions — how to phrase questions Orion can answer.
• Chart types — the full set of charts Celiq supports.
• Mosaics — build dashboards from tiles, including ones pinned from Orion.