One shortcut to everything

Press ⌘K (or Ctrl+K) anywhere in Dema and a single search box opens. Type a few letters and go straight to the Dashboard, Report, Agent, App, MMM, Experiment, or Segmentation set you had in mind. No hunting through the navigation, no remembering where something lives.

It does more than find things. Start a new dashboard, report, or agent session from the same box, invite a team member, or act on the page you are already on: favourite it, duplicate it, copy its link. Recent items sit at the top, so whatever you were just working on is one keystroke away.

Then there is Ask Dema. Type a question, press Tab, and the palette hands it to the Agent along with the page you were looking at, attached as a real reference. The Agent knows you mean that dashboard. Asking about what is in front of you stops being a setup exercise.

Know when an agent needs you

An agent that stops to ask a question or request an approval can now reach you with a desktop notification, so a long run no longer means watching the tab. The notification names the agent and the session, and clicking it takes you straight there.

Permission is yours to give. The first time you open the agents area, a toast offers to turn notifications on, and nothing is requested unless you choose Enable. Notifications are held back while the conversation is already on screen in front of you.

The tab icon carries the same signal without any permission at all. A dot on the Dema mark turns blue while a run is streaming, orange when it needs your input, red on a failure, and green for a run that finished while you were away. 

Bring Excel files into a conversation

You can now attach .xls and .xlsx files to an agent conversation. Spreadsheets go in through the upload button, drag and drop, or a paste, the same way CSV and text files already did.

The agent's Configuration tab has been redesigned to have a better overview on how all of your resources are connected with each other. Resources gathers integrations, apps, skills, files, and any tool permission overrides into rows of chips, with the add and upload actions sitting in the rows themselves. Instructions keeps its own section below.

Files have been improved with Memory capabilities that the agent uses across different runs bringing even more power into your workflows. The file tree view has also been improved to see how the agent structures files and to dig deeper into their content.

We have also improved the UX when removing resources. Now it tells you what resources are affected by change. Disable or delete a skill, delete an app, or disconnect an integration from your connected accounts, and the confirmation lists the agents that will lose access. Disabling a skill that nothing depends on still goes straight through.

Manage allowed models within your organization

You can now grant model access per person, so a team member only sees the models you want them working with. You grant it by tier rather than by model: Frontier for the most capable models at the highest credit cost, Core for balanced day-to-day work, and Light for fast, low-cost models.
It sits on the Credits page under Settings, in the Per-user access table, which now carries a Model access column next to the existing monthly limit.
We have also improved the on-demand credits section with the rate itself, quoted per 1,000 credits. It shows whether or not on-demand is switched on, so you can weigh the cost first.

You can now plug any Model Context Protocol (MCP) server into a custom agent without waiting for it to appear in Dema's integration catalogue. A Custom MCP connector sits pinned at the top of the Add integration dialog, available even when your search finds nothing. Enter the server's URL, choose None or API Key auth, and it joins the agent's toolset like any first-party integration.

You can edit an existing connection in place from the installed card, rotating a key or moving the server to a new URL without disconnecting first.

Push customer segments to Voyado

Voyado joins Klaviyo and Shopify as a destination for customer segmentation. The segments you build in Dema arrive as Voyado labels on a daily cadence, so the tiers and lifecycle definitions you set here become something you can target campaigns with, build segments from, or use to include and exclude contacts in Automations. You will find Voyado under Manage integrations on any customer segmentation set.

Run one incrementality experiment across multiple storefronts

Incrementality experiments now support multi-storefront scopes. When an experiment covers several storefronts, created through the Dema Agent or the API, the details panel lists each storefront on its own line and campaign statistics aggregate across all of them.

Work with files, without the friction

Mention any file or integration inline with @, browse all of an agent's files in one view with indicators for new and changed files, upload .zip archives, and download attachments with a single click.

Agents on less common cadences (monthly, several days a week, or every few hours) now show their real schedule everywhere, and tool approvals default to Always allow, with Allow once in the dropdown, so repeated approvals no longer interrupt your flow.

Agents remember across runs

Each agent now has its own durable, private storage that persists across runs and conversations, so a scheduled or repeated agent can keep lookup tables, cached data, and working notes between sessions instead of starting fresh. You can see what an agent has stored in the workspace tab inside a session, or in the agent's configuration. Runs can also continue for up to five hours, so longer pieces of work have room to finish.

A rough guide to what the Agent costs

We've run fresh benchmarks on typical requests. Sonnet is generally cheaper than Opus for most text work. To help you get a feel for credit usage, here's a rough guide per action.

Map warehouses to storefronts

Warehouse-to-storefront mapping is now available to everyone in Settings, so you can align inventory locations with the storefronts they serve.

Other additions

• Agent can now read and adjust your Causal Factor Attribution (CFA) configuration.

• Sonnet 5 is now available as a faster, lower-cost option for most text work.

• Files the Agent creates while working now show up in its file list automatically at the end of a turn, instead of only being reachable mid-conversation.

• Question sets from the Agent can be collapsed, so a long set of options no longer crowds out the conversation.

You can now queue follow-up messages while the Agent is streaming. Type while it's generating, and the message lands in a queue panel next to the composer, draining one item at a time when the turn finishes. Send now pushes a queued message straight into the running turn so that you can course-correct without cancelling. Queued items can be edited, paused, or deleted, and this feature is available on every custom agent.

A cleaner way to add and share integrations

Integration setup on custom agents has been reworked around a single searchable list. Admins can enable team-wide integrations without leaving the page, and you can attach a connection you've already made personally with one click, instead of another authorization round trip. Widening an agent's visibility now shows exactly which of your connections the new audience would run against, and each credential dialog links to the docs assistant pre-filled with a setup question. Microsoft Teams, OneDrive, and SharePoint join the catalogue. The full, up-to-date list lives in our integrations overview.

Model quality diagnostics for geolift experiments

Incrementality experiments now include a model-quality view alongside business metrics, so you can assess how much to trust a result before acting on it. Instead of taking the numbers at face value, you can see how well the model fits your data, how stable the result stays over time, and how likely the test is to raise a false alarm. If anything is unclear, ask the Dema Agent, and it will walk you through what the diagnostics mean for your experiment. The toggle sits at the top of the experiment statistics panel, and the full walkthrough lives in our incrementality testing guide.

UTM mapping is easier to work with

Existing mappings now sort alphabetically by channel group, channel, and utm_source, with a search field that filters across every UTM field and channel name. Unmapped traffic is ordered by visits, so the biggest gaps sit at the top, and auto-refreshes every 15 seconds. You no longer need to reload to see the effect of a change.

Connect more of your tools

Your agents can now work across more of your external systems and connect multiple accounts for the same integration, which is handy if you run separate stores, regions, or brands. You can also link an app to an agent so it always has the right context, without having to point it there every time. We've added Google Analytics, Google Merchant Center, Gorgias, Contentful, Intelligems, Attentive, Ingrid, Omertia, and Typeform, with more on the way. The full, up-to-date list lives in our integrations overview.

A smoother, sharper agent

Long conversations now stay smooth even when the agent is working through a lot at once, and the plan stays in view the whole time. We've reworked how the agent approaches problems, so it's better at reasoning through tricky, multi-step requests. Follow-up questions appear right above where you type and can be answered straight from the keyboard, and approval requests now lay out clearly what the agent wants to do before acting. We've also refreshed how the agent communicates, so its responses feel clearer and more to the point.

See where your credits go

The Credits page is now always available, so you can keep an eye on your usage at any time. You can filter by person or by agent to see who and what is using credits, both as a breakdown and over time.

A guided onboarding checklist for new merchants

New merchants now get a guided checklist covering the key setup steps. A panel in the bottom-left tracks your progress and takes you straight to each step, then tucks itself away once you're done.

Choose which campaigns to test for incrementality

Incrementality experiments now let you pick exactly which campaigns to include in a test, as well as which to leave out, so you can focus a test on the campaigns that matter. You can also make edits after they're set up.

Your agents, apps, and sessions now live in the side panel, so you can jump straight to what you need from anywhere instead of going through the overview page. Each one has a consistent inline actions menu (rename, configure, copy link, open in a new tab), clearer run statuses, and app tabs that show the app's own name.

Sharing is simpler too. Share an agent or app in a couple of clicks and choose whether it stays private or visible to the team. We confirm before any access change, so nothing is exposed by accident. If you're new to building, our guide on creating agents and apps walks through how to create one, set up data to refresh, and schedule an agent to run it.

You can also view and update your cost and return taxonomies directly through your agent, one of your most-requested features. And under the hood, we've sharpened the specialized flows for MMM and incrementality tests, fixed rendering issues and lag in longer chats, and resolved a Klaviyo connection issue on multi-account setups.

You can now build custom agents in the Dema Agent workspace. Describe the workflow you need, equip the agent with skills, integrations, and reference files, and it handles the work on a schedule you set.

Before taking any action outside Dema, the agent pauses for your approval. You see exactly what it intends to do and either confirm or steer it. This applies across the tools your team already uses, from ad platforms like Meta, Google Ads, and TikTok to Slack, BigQuery, and more.

Agents can build Apps on your data

When an answer needs more than a chart or table, the agent can build an App: a small interactive tool like a budget rebalancer, cohort explorer, or markdown planner. You interact with the App directly, change inputs, filter, and export. The data stays live, so the App keeps working as your numbers update.

See everything the agent produced

Every conversation now has a workspace view that collects what the agent built and did: charts, reports, files, and a log of every action taken in each integration. The agent's current plan is listed alongside, so you always know what it did and what it intends to do next.