BUILD QUERIES WITH THINKY
Working with Thinky
Goal: Turn a plain-English question into a saved warm query, without writing KQL.
Thinky translates what you describe into KQL. The clearer you are about the metric, the time range, and the grouping, the better the first draft. In the Warm Query Manager, Thinky drafts the query into the window; nothing is saved until you click the canvas Save icon.
THE LOOP
Describe, preview, save
You describe the answer you want. Thinky drafts a query. You run a preview. You save it, refine it, or throw it away. Saved queries become a shared source of truth; discarded drafts cost nothing.
Four asks that cover most needs
| You ask | Thinky drafts (in KQL) |
|---|---|
| "Average signups per hour over the last day" | a summarize count() by bin(Timestamp, 1h) |
| "Every hour with more than 100 signups" | a where threshold over hourly buckets |
| "Signups by campaign this week" | a metric split by Campaign |
| "Unusual spikes in conversion this week vs last" | a multi-step query with a baseline and comparison |
Tips for a better first draft
- Name the time range ("last 24 hours", "this week").
- Name the fields you mean ("by campaign", "by source").
- Say the aggregation ("average", "count", "sum").
- Say the order if it matters ("highest first").
Refining a draft
| Say | Thinky does |
|---|---|
| "Group by day instead of hour" | swaps bin(Timestamp, 1h) to 1d |
| "Only the spring campaign" | adds where Campaign == "spring" |
| "Raise the threshold to 250" | edits the where value |
| "Sort newest first" | flips order by to desc |
When to edit the KQL yourself
For a small tweak - a number, a field name, a sort direction - it's often fastest to edit the query text directly. KQL Basics is all you need to read first.
Tell Thinky how your team works
Put the rules every AI should follow in the workspace's Instructions, in the workspace settings: currency, naming, what not to touch. Thinky reads them on every turn. When you correct it with a rule like that, it may offer to add the line for you to accept. See Workspace instructions.
You stay in the loop
Thinky drafts; you decide. From any external AI over MCP it can test ideas as drafts (call_api) and
propose the saved query (manage_proposals), but accepting always happens in the workspace - there is no
accept tool over MCP, by design. Accepting is not written to
Provenance today.
Greenfield: when you want to go faster
The main Thinky panel's settings cog holds its mode. In Greenfield Mode the main Thinky's own proposals - a new warm query among them - are accepted for you as soon as Thinky makes them, with your rights. Greenfield never deploys, some changes still wait for a person, and agents connected over MCP still only propose. A warm query accepted this way serves at its address after the next Deploy, like any accepted warm query. The Thinky in the Warm Query Manager never proposes, so there you always save the draft yourself. See Greenfield.
Next steps
- Read or tweak the KQL -> KQL Basics
- Save and reuse it -> Warm Queries
- The boundary, in one page -> You stay in the loop
- Let Thinky accept its own proposals -> Greenfield
On fathym.com: OpenX →