add_card_to_dashboard
ChatGPTMUTATES DATA: attaches an existing chart from the site's library to a dashboard (max 15 charts per dashboard). The chart itself is unchanged and can sit on several dashboards at once. Reversible with remove_card_from_dashboard.
aggregate_sessions
ChatGPTCount or average sessions grouped by a dimension (country, device, page…) without listing them one by one, optionally compared with the previous period. Use for questions like 'which countries had the most errors this week vs last week'.
compare_pages_by_frustration
ChatGPTRanks every page on a site by rage, dead, or error clicks so you can find the highest-friction pages without checking each one. Follow up with get_page_heatmap (clicks mode) on a page to see which elements cause the frustration. Use for questions like 'which pages have the most rage clicks this week?'
create_card
ChatGPTMUTATES DATA: saves a new chart (card) to the site's chart library, and optionally places it on a dashboard in one step. Preview the spec first with the matching read tool for this chartType (get_metric/get_trend/get_breakdown/get_page_ranking/get_feedback_snapshot/get_page_flow), and confirm with the user before saving. querySpec uses a measure with exact names (e.g. 'totalSessions', not 'sessions') plus a groupBy for breakdown/table charts. When attaching to a dashboard, leave layout out to auto-place (recommended). The grid is 3 columns: x=0–2, w=1–3 with x+w<=3, h always 1.
create_dashboard
ChatGPTMUTATES DATA: creates a dashboard and optionally fills it in one step — with brand-new charts (newCharts) and/or existing library charts (existingChartIds). Max 15 charts. Confirm the plan with the user before calling. Each chart needs a chartType and a querySpec. querySpec uses a measure with exact names (e.g. 'totalSessions', not 'sessions') and, for breakdown/table charts, a groupBy (e.g. ['country']). Layout: the dashboard is a 3-column grid — columns are x=0,1,2; width w is 1–3 with x+w<=3; height h is always 1. Simplest and recommended: leave layout out on every chart and they auto-place left-to-right, top-to-bottom. Titles: chart titles have no time window (no '(7d)', 'last 30 days', etc.) — cards carry no fixed time filter; the window is applied when the dashboard is viewed.
create_funnel
ChatGPTMUTATES DATA: saves a new ordered-step conversion funnel on a site and kicks off its initial processing (unless it's a multi-session funnel). Build steps with generate_funnel_steps_from_prompt first and confirm the step order with the user — this tool only shape-checks steps, it cannot tell a correct funnel from a wrong one. A funnel is not a segment; create_segment doesn't create funnels.
create_heatmap
ChatGPTMUTATES DATA: saves a page/device/segment/time-range combination as a named heatmap view in the dashboard. Confirm the page path and audience with the user before calling.
create_segment
ChatGPTMUTATES DATA: saves a new named audience segment on a site. Build the filters with generate_segment_from_prompt first — do not hand-write them. This tool runs the real filter → query conversion before saving and rejects filters that don't actually resolve to anything. Confirm the segment name and audience with the user before calling.
delete_card
ChatGPTMUTATES DATA — DESTRUCTIVE: soft-deletes a chart from the site's library, removing it from every dashboard that shows it. Only call after the user has explicitly confirmed deleting this specific chart by name.
delete_dashboard
ChatGPTMUTATES DATA — DESTRUCTIVE: soft-deletes a dashboard. The charts on it stay in the site's chart library (they are not deleted). Only call after the user has explicitly confirmed deleting this specific dashboard by name.
delete_funnel
ChatGPTMUTATES DATA — DESTRUCTIVE: permanently deletes a funnel and its precomputed results. Only call after the user has explicitly confirmed deleting this specific funnel by name.
delete_heatmap
ChatGPTMUTATES DATA — DESTRUCTIVE: permanently deletes a saved heatmap view. The underlying session data is untouched; only the saved page/device/segment/time-range bookmark is removed. Only call after the user has explicitly confirmed deleting this specific heatmap by name.
delete_segment
ChatGPTMUTATES DATA — DESTRUCTIVE: permanently deletes a segment. Any alert, dashboard, or Lift AI goal built on it will lose that reference. Only call after the user has explicitly confirmed deleting this specific segment by name.
generate_funnel_steps_from_prompt
ChatGPTDraft a funnel's ordered steps from a plain-language description (e.g. 'home page, then pricing, then signup'), using the same generator as the FullSession dashboard. Uses an AI model to draft it. Nothing is saved; show the returned stepNames to the user to confirm the sequence. This is the way to build funnel steps for create_funnel/update_funnel (the steps format is deeply nested to write by hand); pass steps through unchanged.
generate_segment_from_prompt
ChatGPTDraft a segment filter from a plain-language audience description (e.g. 'mobile visitors from Germany who added to cart'), using the site's custom events and attributes. Uses an AI model to draft it. Nothing is saved; the user can apply it in the FullSession segment builder.
get_ai_recommendations
ChatGPTThe friction issues Lift AI found for a goal — what to fix, why, where (page + funnel steps), predicted lift, and the evidence 2×2 session counts — plus the latest run's conversion prediction inline. Pass a recommendation's id to get_recommendation_sessions for the underlying replays. Use for questions like 'what should we fix to improve conversions?'
get_breakdown
ChatGPTComputes an ad-hoc ranking — a measure grouped by a dimension (country, page, device, error message, etc.) — the same way a breakdown card on a dashboard would. Nothing is saved. querySpec.groupBy is required. Example: querySpec={measure:'rageClicks', groupBy:['page']}. Use for questions like 'which countries bring the most traffic?'
get_dashboard
ChatGPTA dashboard's full definition: its charts (cards) with chart type, querySpec (measure/groupBy), and grid layout. Use get_dashboard_data to compute the actual numbers.
get_dashboard_data
ChatGPTRuns every chart on a dashboard for a time window and returns the computed series — the same numbers the dashboard shows in the app. Each result is keyed by chart id and includes the chart's title for context.
get_element_interactions
ChatGPTAggregated click/frustration stats for a single element (by CSS selector or visible text) across all sessions on a page, plus the sessions where it was rage/dead/error clicked with replay URLs. Use get_page_heatmap (clicks mode) first to discover element selectors.
get_feedback_snapshot
ChatGPTComputes an ad-hoc feedback rollup — current sentiment/scores per widget, or the same over time — the same way a feedback card on a dashboard would. Nothing is saved. Example: querySpec={feedbackView:'overview'}.
get_form_analytics
ChatGPTPer-form interaction stats aggregated across sessions: impressions, completion rate, and abandonment rate, with the pages each form appears on. Note: stats are per form element — field-level data isn't available. To watch users abandon a specific form, follow up with search_sessions using filters.hasAbandonedForm.
get_funnel_analysis
ChatGPTStep-by-step conversion for a funnel: users per step, drop-off, median time and interactions between steps, plus overall entrants/finishers/conversion rate. Optionally break down by country/device/browser/os, or compare saved segments against the funnel's audience. Use for questions like 'where do users drop off in the checkout funnel?' The user sees a funnel card with breakdown tabs (device/country/browser/os), so call this once; only call again with a breakdown if the user explicitly asks about one.
get_funnel_sessions
ChatGPTPaginated session recordings of users who dropped at (or converted through) a funnel step, with replay URLs — the "show me who got stuck at step 2" tool. Run get_funnel_analysis first to see where the drop-off is.
get_heatmaps
ChatGPTSaved heatmap view definitions (page/device/segment/time-range combinations users bookmarked in the dashboard). Use get_page_heatmap with the same page/device/segment/window to compute the actual data. Use create_heatmap/update_heatmap/delete_heatmap to manage them.
get_metric
ChatGPTComputes one ad-hoc number — a total or average for a measure over a time window — the same way a metric card on a dashboard would. Nothing is saved. Example: querySpec={measure:'totalErrors'}.
get_page_flow
ChatGPTComputes an ad-hoc Sankey of how users move between pages — the same way a sankey card on a dashboard would. Nothing is saved. nodes and edges come back with short generated ids and edges already resolved — use them as returned, never regenerate ids or rewrite the edge list yourself. Example: querySpec={maxSteps:5, topNodesPerStep:5}.
get_page_heatmap
ChatGPTAggregated interaction data for one page across all matching sessions, in one of three modes: clicks (per-element click/rage/dead/error breakdown ranked by frustration), scroll (how far down the page users scroll), or movement (mouse-movement density zones and hotspots). Use get_element_interactions to deep-dive one element from the clicks result. Use for questions like 'what do people click on the pricing page?' or 'how far do visitors scroll on /blog?'
get_page_ranking
ChatGPTComputes an ad-hoc page ranking — most-viewed or slowest pages, by visited/landing/exit role — the same way a pages card on a dashboard would. Nothing is saved. Example: querySpec={pageRanking:'mostViewed', pageKind:'landingPage'}.
get_recommendation_sessions
ChatGPTPaginated session recordings from a recommendation's evidence, split by whether the user hit the issue (affected/notAffected) and whether they converted — with replay URLs. E.g. "users who hit this issue but still converted" = type affected + outcome converted.
get_segments
ChatGPTList the saved audience segments on a site (named visitor filters, e.g. 'Mobile users in the US'), with the fields each one filters on. Segments with usableAsSegmentId=true can be passed as segmentId to get_page_heatmap so results match the dashboard for that audience. Use create_segment/update_segment/delete_segment to manage them.
get_session
ChatGPTGet the details of one session: metadata, a summary of its signals, and a step-by-step timeline of what the visitor did, plus a replay link. Optional extra sections: pages, errors, clicks, network, forms. Use to explain what happened in a specific session.
get_site_overview
ChatGPTOne-call health summary for a site over a time window: total sessions, unique visitors, average duration and page count, frustration totals (rage/dead/error clicks), and top landing pages, countries and devices. Start here before drilling into search_sessions, heatmaps or funnels.
get_trend
ChatGPTComputes an ad-hoc trend — a measure's daily values over a time window — the same way a timeseries card on a dashboard would. Nothing is saved. Add querySpec.groupBy to split the trend into one line per value of a dimension instead of a single overall line — that is what answers "sessions by device over time", "daily errors per browser", "traffic per country over the last month". This tool does that in one call, without session-level aggregation. Use querySpec.series to scope the chart instead — one entry filters the whole chart, several give one named line each. That is how you build a separate chart per device: call this once per device with series=[{name:'Desktop', filters:[{field:'device', value:'desktop'}]}] and groupBy:['country']. Splitting one combined result by hand breaks the alignment between each series and the timestamps, so request each chart separately. The response returns timestamps plus series: [{name, data}] with one entry per dimension value, every data array the same length as timestamps and zero-filled. Use them exactly as returned — never rebuild, pad, trim, re-pivot or re-derive them. Use for questions like 'how did sessions change over the last 30 days?' Examples: querySpec={measure:'totalSessions', comparePreviousPeriod:true}; one line per device without groupBy: querySpec={measure:'totalSessions', series:[{name:'Desktop',filters:[{field:'device',value:'desktop'}]},{name:'Mobile',filters:[{field:'device',value:'mobile'}]}]}.
get_user_sessions
ChatGPTProfile and visit history of one identified visitor (by userId or email): first and last seen, devices, top countries, latest custom fields, and every session with replay links. Use when the user asks about a specific customer or visitor.
list_ai_goals
ChatGPTLift AI is FullSession's AI analysis that finds what blocks conversions in a funnel. Lists Lift AI goals (funnel + KPI targets the AI analyzes for friction) with each goal's latest run prediction. The entry point before get_ai_recommendations — use it to find the goalId.
list_cards
ChatGPTThe site's chart library: every saved chart/card with its type, measure/groupBy querySpec, and description — whether or not it sits on a dashboard. Use get_metric/get_trend/get_breakdown/get_page_ranking/get_feedback_snapshot/get_page_flow to compute a chart definition ad hoc. Use add_card_to_dashboard to place a saved one.
list_dashboards
ChatGPTThe dashboards visible to the authenticated user on a site, with their charts (cards) and grid layouts. Pass a dashboard id to get_dashboard_data to compute its numbers.
list_funnels
ChatGPTConversion funnels defined on a site with their step names, base segment and status. The entry point before get_funnel_analysis / get_funnel_sessions — use it to find the funnelId.
list_sites
ChatGPTList the websites (sites) the connected account can access, with each site's ID and URL. Call this first when the user hasn't said which site they mean, or to match a site name or URL to its siteId.
remove_card_from_dashboard
ChatGPTMUTATES DATA: detaches a chart from a dashboard. The chart stays in the site's library and on any other dashboards — this only removes the placement. The inverse of add_card_to_dashboard.
search_sessions
ChatGPTFind session recordings on a site that match filters (time window, user, page, country, device, frustration signals…). Returns short summaries with frustration counts, a replay link, and a sessionId for get_session. Use for questions like 'show me sessions with rage clicks on /checkout yesterday'. signals.rageClickBursts counts bursts of rapid clicking (one burst is several clicks), so it is lower than the individual rage-click totals from get_site_overview and the heatmap tools.
search_site_knowledge
ChatGPTSearches the site's crawled, indexed knowledge sources (documentation, marketing pages, pricing, support content, etc. — declared via the Daleel Knowledge sources screen) for passages relevant to a query. Use this when a question is about what the product does, how it works, pricing, or anything else better answered from the site's own declared content than from session/analytics data. Each result carries its source URL — cite it when you use the content.
update_card
ChatGPTMUTATES DATA: updates a saved chart's title, description, type, or querySpec. Only the provided fields change; querySpec, when given, replaces the whole spec (preview it first with the matching read tool for this chartType — get_metric/get_trend/get_breakdown/get_page_ranking/get_feedback_snapshot/get_page_flow). The change affects every dashboard showing this chart. Confirm with the user before calling.
update_dashboard
ChatGPTMUTATES DATA: updates a dashboard's name, description, or privacy. Only the provided fields change. Confirm with the user before calling.
update_funnel
ChatGPTMUTATES DATA: changes a funnel's name, base audience, steps or chart settings. Only the fields you pass change; everything else keeps its current value (to rename, pass just funnelId and funnelName). Changing steps replaces all of them and reprocesses the funnel's data. A funnel can't be edited while it is still processing (a few minutes after creating it or changing its steps). Confirm the change with the user before calling.
update_heatmap
ChatGPTMUTATES DATA: changes one or more fields of a saved heatmap view (title, page, device, segment, time range, funnel scope). Unlike segments/funnels this is a partial patch — omitted fields are left as-is. Confirm the change with the user before calling.
update_segment
ChatGPTMUTATES DATA: changes a segment's title, description or filters. Only the fields you pass change; everything else keeps its current value (to rename, pass just segmentId and title). Changing filters replaces all of them; build new filters with generate_segment_from_prompt, not by hand — this tool runs the real filter → query conversion before saving and rejects filters that don't actually resolve to anything. Confirm the change with the user before calling.
whoami
ChatGPTShow which FullSession account is connected: the signed-in user's email, customer ID and accessible site IDs. Use when the user asks which account or sites they're connected with.