Skip to main content
​
FeatureFix
Several comparisons on one KPI tileA KPI tile can now carry up to three deltas instead of one, each on its own labelled row — so the same revenue figure can be read against last year, last month and target without repeating the headline number in three separate tiles. Each delta keeps its own good/bad colouring, and comparisons the source data has no column for are left off rather than invented. Ask for the baselines you want by name: “revenue, sessions and CVR, each with the change vs. last year and vs. target”. KPI tiles in reports →Fixes
  • The data summary table in chat no longer logs a React key warning while a tool call is still streaming in
​
FeatureImprovementFix
Two new chart typesAgents can now build ranked bar charts — horizontal bars ranked by value, so top-N comparisons keep their category names readable instead of rotating them under a vertical axis — and scatter charts for the relationship between two numeric metrics, with optional bubble size, a colour per group, point labels, and a least-squares trendline with its R². Time series charts gained cleaner vertical spacing and padding throughout. Which visualisation to use, and how to ask for it →KPI tiles in reportsReports can now open with a proper scorecard row instead of a wall of text: up to nine KPI tiles in a three-column grid, each with a label, a formatted headline value, and an optional change vs. a comparison period. Deltas are coloured by whether the movement is good or bad for the business rather than by direction alone — so a rising CPA reads red and a rising conversion rate reads green.Follow the agent while it worksAgent reasoning is now visible in chat as a collapsible Thinking block (Thought for Ns once complete), and is kept with the message so it’s still there when you reopen the chat. A stop button lets you end a running turn while keeping whatever the agent already produced. Turns also survive a dropped connection — close your laptop or lose Wi-Fi and the agent keeps working server-side, with the chat reconnecting and catching up on anything it finished while you were away.More reliable charts and reportsChart configurations are now validated against the actual data before rendering, with errors the agent can act on rather than a broken chart. Long multi-section reports no longer fail wholesale on a single malformed block — a common structural mistake is repaired automatically instead of discarding the whole report. Agents also read row limits on query results correctly now, so they no longer chase data they already have or describe a chart built from thousands of rows as “only a few rows were available”.Teamwork boardsOrganizations can now enable Teamwork, connecting a board to your project tasks and optionally restricting access to specific user groups. Once enabled, users can view board columns, create and edit cards (title, description, assignee, epic), drag cards between columns to move work through a workflow, and leave comments — all kept in sync with your project management tool.Longer conversations without losing contextChats approaching a model’s context limit are now compacted automatically in the background — older turns are folded into a summary while the underlying data and tool outputs stay fully available, so long-running conversations keep working smoothly. A manual Compact conversation option is also available from the context-usage indicator, which now shows how much of the context window is currently in use separately from total tokens billed.More reliable reportsReport generation now automatically retries on transient failures instead of silently producing a broken report.Claude Sonnet 5 supportClaude Sonnet 5 is now available as a model for agents, with pricing configured. We also fixed a missing price for Claude Opus 4.7 and corrected a default token limit that had been set far too low.Smaller improvements
  • Docs and Changelog links added to the sidebar, with improved tooltips when the sidebar is collapsed
  • The Settings shortcut now shows consistently for organization admins in the chat sidebar
  • The chat API now returns charts and tables alongside reports, not just legacy report documents
​
FeatureImprovementFix
Attribution Insights refreshThe Attribution Insights comparison view is now a sortable table with Share, Revenue, and Margin view modes per channel, plus new spend, ROAS, and POAS columns alongside Markov, Shapley, and Last-touch deltas.MMM Scenario Planner (beta)A new Scenario Planner lets you interactively re-allocate marketing budget across channels using sliders constrained to each channel’s modeled range, and see projected response curves, an allocation table, marginal-ROI charts, and a live KPI summary update — without waiting on a new model run. MMM channel cards also show more channels, and empty states are clearer when a label has no modeled data.Clearer approval flowThe job-approval banner was redesigned to clearly show whether your approval is required or pending, along with an expiry time and Approve/Reject actions. Fixed an issue where rejected jobs could show an incorrect badge in chat history.Agent configuration improvements
  • Attach named example queries to a database agent to steer it toward known-good query patterns
  • Flag an agent’s answer as incorrect, with a reason, for admins to review
  • Agent answers can now show a small panel listing the data source, tables, and filters used
  • Attach notes to a specific database table to help agents interpret it correctly
Smaller improvements
  • Notification channel filters reworked into a table with Select all / Deselect all shortcuts
  • Scheduled job emails now show the resolved run name instead of the static job template name
  • Reorganized the organization settings page into clearer sections
​
FeatureImprovement
MMM Model ResultsMedia Mix Modeling now has a full Model Results view: a channel contribution breakdown, fit-vs-actual series, response curves, adstock decay and saturation curves, marginal-ROI charts, and a model health panel — everything needed to understand what a model run found, in one place.Attribution table improvementsAttribution overview tables now show spend alongside each channel’s contribution, and you can optionally include margin as a comparison metric next to revenue.Data Book panelThe Media Mix Modeling data book was reworked into its own panel, with separate input and final-table views and clearer handling of sources like direct uploads and MMM platform data.Approval banner for job runsJobs that require approval before running (including fanned-out/parameterized runs) now show a dedicated approval banner in the chat, so it’s clear when a run is waiting on you.
​
FeatureImprovementFix
Personal follow-up chat for job runsJob run chats (scheduled and API-triggered) are now explicitly read-only — they represent a fixed, reproducible snapshot of what the agent returned. To continue the conversation from a job result, use the new Start Follow-up Chat button. This creates a personal copy of the chat, pre-loaded with the original messages, that you can continue freely without affecting the source run.If you’ve already created a follow-up for a run, a Go to follow-up chat button appears so you can pick up where you left off. A Show source link in the follow-up header takes you back to the original run at any time.Tool panel redesignThe tool activity panel has been redesigned — tools are now more clearly named, each tool shows an info popover explaining what it does, and the overall layout is cleaner and easier to scan during a running conversation.Test notification channelsYou can now test a notification channel directly from the settings page before connecting it to a job. Send a test message to confirm your Slack, Teams, or email channel is working correctly.Report reliability improvementsSeveral fixes to report generation: better error handling when the report tool encounters unexpected output, a fallback path to prevent silent failures, and a fix ensuring newly generated V2 reports are immediately downloadable.
​
FeatureImprovement
Scheduled PDF reportsAgents can now produce fully formatted, PDF-exportable reports as part of a scheduled job. Include a report instruction in your job prompt and each run delivers a structured report — with sections, tables, and charts — ready to share or export.Parameterised jobsJobs now support parameters. Define a variable (like market or campaign_id) and run the same job across multiple values at once. Each value gets its own result. You can also choose to run just one combination instead of the whole set.Smarter agentsThe Data Analysis Agent has been rebuilt from the ground up — faster responses, better reasoning, and more accurate analysis output. Agents also now remember what they’ve learned about your data between runs, so you don’t need to repeat context every time.Chat improvements
  • New message filter — quickly find user messages, agent responses, or tool activity in a long chat
  • Redesigned chat layout with a cleaner panel design and better readability
  • Agents now open on your most recently used agent after logging in
Notification channels movedYou’ll now find notification channels under Settings → Notifications — previously they were in a different part of the settings.
​
FeatureImprovement
New chart engineTimeseries charts have a new look — cleaner, more interactive, and easier to read. Charts can also now be embedded directly inside agent-generated reports.Table artifactsAgents can now produce structured data tables as standalone outputs alongside charts — useful when you want the raw numbers without a chart.Learning between sessionsAgents now save and recall observations about your data across sessions. If an agent learns something helpful about a dataset or metric, it’ll remember it the next time you ask.Smoother chat performanceGeneral chat performance improvements — less flickering, faster message rendering, and a more stable sidebar.