Cognis chat
Ask questions about your BigQuery cost in plain English. Cognis answers from your own scan data. On Starter, Growth, Pro and Enterprise.
Cognis is the assistant inside SlashPeak. Open it from Cognis in the sidebar (/cognis). Ask a question in plain English. Cognis looks up the answer in your scan data and replies in words, tables and numbers.
Cognis is not a general chatbot. It only talks about your warehouse cost and usage.
| Plan | AI usage per month |
|---|---|
| 30-day trial | None (no AI in the trial) |
| Starter | 1x |
| Growth | 2.5x |
| Pro | 5x |
| Enterprise | Custom |
Cognis chat, the AI query optimizer and the AI DBA share this allowance. See Plan limits for how it is counted.
How Cognis answers
For each question, Cognis does two steps:
- Look up. It picks one lookup that answers the question. It prefers the same metrics the dashboards use (spend, top queries, storage, findings, reservations and so on), so the numbers match. If none fits, it writes one read-only query over the scan data SlashPeak keeps for your account. If that query fails, it gets one repair attempt.
- Answer. It writes the reply from that data only. It does not invent dollar amounts, table names or users. If nothing matched, it says so.
Times and dates in the reply use your account timezone.
What to ask
A new chat shows three starter questions. More appear while Cognis works. Some examples:
- Where is the waste this month?
- Which 5 queries cost the most this week?
- Which users spent the most on queries this month?
- Which tables have not been queried in the last 90 days?
- What are the top open findings by monthly savings?
- Compare compute spend this week with last week. Which day changed the most?
- Which datasets hold the most long-term storage, and would physical billing be cheaper?
- Is our reservation oversized or contended?
- Should this project stay on-demand or move to Editions?
- At which hour of the day do we spend the most on queries, and which users drive it?
What Cognis can see
- Job metadata from your scans: cost, bytes, slots, users, query patterns, failures.
- Table and dataset catalog: sizes, storage cost, last queried time.
- Findings and their savings.
- Reservations and slot use.
Cognis cannot see table rows. It cannot change BigQuery, IAM or your settings. To have fixes made for you, see Cognis AI DBA.
Chats
- New chat starts a fresh conversation.
- The chat list on the left keeps your old chats. You can pin, rename, archive or delete each one.
- Each chat gets a short title from its first question.
- A very long chat may hit the size limit. Start a new chat if Cognis asks you to.
AI filters on Usage
The filter bar on Usage has a Cognis box: "Describe the jobs you want. Cognis picks the filters for you." Type, for example, "service accounts running dbt in prod". Cognis turns it into filter conditions you can see and edit. This is on the same plans as Cognis chat.
The query optimizer
On the dashboard Top queries chart and on Usage → Query, hover a query and click Optimize. Cognis opens in a new tab and suggests a cheaper rewrite. See AI query optimizer.
Activity
Cognis → Activity starts with the Plan usage card: how much of this month's AI usage is used, and the reset date. On Growth, Pro and Enterprise, it also shows the fix pull requests open now and the GitHub repos connected against your plan, and the dbt Cloud CI runs each pull request gets. Below it, the log lists every AI DBA step, every AI query optimizer call and changes to AI settings, newest first. Owners use it to see who did what and when. Chat questions and answers are not in this log. They stay in the chat list.
Privacy
Cognis chat runs on Google Gemini, on Google Cloud Vertex AI. Each question sends Gemini your question, a few earlier messages of the chat and the rows Cognis looked up for it. If Gemini's first read-only query fails, Claude (Anthropic's model, also on Vertex AI) fixes the query. Table contents are never sent, because SlashPeak never reads them. See AI and your data.
If a number looks different
Cognis uses SlashPeak's cost estimate, the same as the dashboard. Your Google invoice can differ (credits, discounts, tax, non-BigQuery products). For invoice totals, see BigQuery billing summary.
Messages you may see
| Message | Meaning |
|---|---|
| "Cognis AI is on paid plans. The trial has no AI." | You are on the trial. Choose a plan in Settings → Billing. |
| "You used all of this month's AI usage on the … plan. It resets on …" | Your plan's monthly AI usage is used up. Wait for the reset date, or upgrade. |
| "The AI service is busy right now" | A short pause on the AI provider side. Try again in a few minutes. |
| "That message is too long or the conversation is too long" | Start a new chat. |