AI chat

AI chat is a panel, not a page. It reads the current workspace and writes nothing to it. Only you can save from it, with a button: a chart to Reports or a dashboard, or an answer as an analysis.

Open it

Select New chat at the top of the sidebar. The panel opens on the right, on a new conversation. If a conversation is open, it moves to Chat history.

Control	What it does
The title	Rename the conversation
Chat history	Reopen a previous conversation
Expand / Collapse	Give the chat the whole page, or put it back beside the page. `⌘⇧E` on a Mac, `Ctrl+Shift+E` elsewhere.
Close chat	Close the panel. The conversation is kept

Type in the box marked Ask about your data.... The word thinking appears while the answer streams. Select the stop control to end a turn early. An untitled conversation reads New chat. We name it from your first message, truncated to 60 characters. When the chat is expanded, New chat is highlighted in the sidebar. Select any page in the sidebar to open it: the chat moves beside the page and keeps the conversation. Select Expand to go back to the full-page chat.

What it can read

Chat holds read tools only, and each one maps to a page you can open yourself.

Tool	Answers
`get_overview`	Visitors, pageviews, sessions for a range
`get_traffic`	Referrers and campaigns
`get_devices`	Browser, device, operating system
`get_locations`	Country and city
`get_events`	Individual event rows
`list_event_properties`	Which custom properties an event carries, with a sample value
`get_event_names`	Which event names exist
`get_realtime`	The last 30 minutes
`get_attribution`	Attributed revenue by channel, source, campaign, ad set, or ad
`get_users`	The user list
`get_user_detail`	One person’s journey
`get_ads`	Spend, clicks, ROAS per platform
`get_revenue`	Revenue from connected sources
`get_commerce_metrics`	Certified revenue, spend, profit, margin, ROAS, MER
`list_metrics`	Which metric ids you can ask for
`query_metrics`	Up to 50 named metrics over a range, with a comparison
`get_retention`	Retention cohorts, period by period
`get_ltv`	Cohort revenue per acquired person
`get_funnel`	An ordered funnel over 2 to 5 event names
`get_tags`	Tracking-link `lyr` values
`get_workspace`	Workspace settings
`get_onboarding_status`	What is set up, and the next steps a person must take
`get_usage`	Event usage this billing period
`search_docs`	Matching pages of the Datalyr documentation
`analytics_query`	Any metric by dimension, filter, and date range
`list_dimensions`	Which dimensions a query can group by
`get_dimension_values`	The exact stored names of campaigns, sources, and events
`get_context`	Timezone, currency, connected sources, and attribution defaults
`search`	The id behind a name you typed
`get_subscriptions`	Active subscriptions, MRR, ARR, and trials ending per day
`list_replay_sessions`	Recorded sessions that match a source, campaign, device, or friction filter. See Session replay
`get_replay_summary`	What happened in up to 20 recorded sessions
`list_reports`	The cards on the Reports page, in board order
`get_report`	One saved report and what it runs
`run_report`	A saved report’s numbers for a range
`list_dashboards`	The workspace’s dashboards
`get_dashboard`	One dashboard, its range, filters, and cards
`run_dashboard`	Every card’s numbers on a dashboard, up to 20 rows each
`list_analyses`	The workspace’s analyses
`get_analysis`	One analysis, with its body and the queries behind it
`list_share_links`	The share links to a report, dashboard, or analysis. Never the URL.
`get_health`	The Health page as data: capture, identity, delivery, match quality, and what needs fixing
`list_deliveries`	Delivery rows one per event and rule, with the reason each one was skipped or failed
`list_unlinked_orders`	Which Shopify orders linked to an ad visit, which did not, and the miss code for each
`get_match_signals`	For one ad platform, which match signals each event type carried, and why a click ID is missing
`list_conversion_rules`	Every conversion rule with its trigger, destination, scope, and this window’s delivery counts
`get_connection_health`	Each connection’s credential state, last sync, and the matching capture and delivery states

Chat has no write tool. It cannot change settings, billing, sources, conversion rules, reports, dashboards, analyses, or share links.

Read an answer

An analytics_query answer appears as a card in the panel.

Part	What it shows
Chart or table	A line chart for a time series, a bar chart for up to 12 values, otherwise a table. Select `+N more` for the rest of the rows.
Totals	One line per metric, with the change against the previous period when you asked for one
`how this was calculated`	The metrics, date range, timezone, filters, and attribution model and window
`open in app`	The page in Datalyr that shows the same numbers
Save to Reports	Adds this chart to the Reports page. See Save a chart to Reports.
Add to dashboard	Appears after you save. Puts the card on a dashboard.

A metric the source could not measure shows its reason instead of a number. It is never shown as 0. Under each answer, select the thumbs up or thumbs down. A thumbs down asks what was wrong: Wrong number, Incomplete, Wrong data or range, Too slow, or Other. We read every one.

Save a chart to Reports

Find the answer card you want to keep.
Select Save to Reports under the chart.
Wait for Saved to Reports · Open.
Select Open to see the card on the Reports page.
Select Add to dashboard to put the card on a dashboard.
Pick a dashboard, or New dashboard.

The button saves the query the chat ran: its metrics, dimension, filters, and attribution model. It does not save the numbers. The card reruns on the Reports page range.

Answer	Saved as
A time series with no dimension	A line card
One dimension, no time buckets	A breakdown card
Totals only	A metric tile
A `get_funnel` answer	A funnel card with the same steps. The window is the chat’s window, or 7 days when it set none.
Two or more dimensions, or a dimension with time buckets	No button. The Reports page cannot draw it.

Reports counts a funnel from raw events and looks back up to 365 days. The chat’s get_funnel looks back 90 days, so the two funnels can show different numbers.

Anything the chat saves can be edited on the Reports page. The model never selects the buttons, and it has no tool that saves. Over the API, a report saved from chat reads created_via: "chat".

Save an answer as an analysis

Wait for the answer to finish.
Select Save as analysis under it, next to the thumbs.
Wait for Saved · Open.
Select Open to read the analysis.

The button appears under an answer that ran at least one analytics_query.

Saved	From
Title	The answer’s first heading, or its first line, up to 120 characters
Body	The answer’s text, as Markdown
Queries	Every `analytics_query` the answer ran, up to 20, under How this was calculated
Dates	The exact dates each query covered. A `last 30 days` question keeps those 30 days.

The analysis shows the same numbers later, because each query keeps its dates. A query the analysis cannot run again is dropped, and the rest are saved. Edit it on the Analyses tab under Reports. See Dashboards and analyses.

Limits

Limit	Value
Model	`claude-sonnet-5`
Tool steps per turn	16
Turn duration	300 seconds
Workspace scope	The active workspace only

Chat cannot read another workspace, whatever a message asks it to do. To ask about a different workspace, switch workspace first.

Ask a question that gets a useful answer

Name the metric or the object.
Give a date range.
Name the platform, campaign, user, or event.
Ask for the rows when a total surprises you.

These prompts each meet the first three.

Which campaigns drove the most attributed revenue in the last 30 days?
How many trials end this week?
Compare Meta and Google spend and ROAS this week.
Show me the journey for user_123.
Which event names fired today?

Answers are generated, so check anything that drives spend before you act on it. Every answer comes from a page you can open: Overview, Events, Users, or Reports.

Verify

Ask Which event names fired today?.
Open Events and set the range to today.
Confirm the event filter offers the same names.

When it does not work

What you see	Cause	Fix
`No chats yet`	You have no conversations	Select New chat
Chat returns no data	The workspace has no agent key	Open Settings → API and create one
An answer stops mid-way	The turn hit 16 tool steps	Ask a narrower question
A turn ends with no answer	The turn hit 300 seconds	Shorten the date range
The answer covers the wrong workspace	The wrong workspace is active	Switch workspace in the sidebar

An agent key starts with dk_agent_. Claude and Codex connect through hosted MCP with OAuth, so you never paste this key into them. Next, read the MCP setup to ask these questions from your own editor.

Get started

Installation

Revenue

Leads

Using Datalyr

Advanced

Integrations

Account & billing

Troubleshooting

Open it

What it can read

Read an answer

Save a chart to Reports

Save an answer as an analysis

Limits

Ask a question that gets a useful answer

Verify

When it does not work

​Open it

​What it can read

​Read an answer

​Save a chart to Reports

​Save an answer as an analysis

​Limits

​Ask a question that gets a useful answer

​Verify

​When it does not work

Open it

What it can read

Read an answer

Save a chart to Reports

Save an answer as an analysis

Limits

Ask a question that gets a useful answer

Verify

When it does not work