// changelog
Every entry is written when the change ships, by the people who shipped it, and it is the same list the product shows inside your dashboard.
// the record
A setting that read like a preference was deciding whether a customer got an answer at all: mail outside the configured language was escalated to a person, and on a desk with nobody behind it that meant silence. The customer's own language now decides the reply, whatever the store's markets are.
The examples the drafter imitates are now taken from what your team really sent, with the quoted history and the mail template cut away first, and per store and per language rather than per fleet. The gap between the two is measured every week, so a desk drifting away from its own register shows up as a number.
Each support thread now has a unique link you can share with a customer, letting them read the full conversation and send a reply that arrives in your inbox exactly like an inbound message. A customer reply will also automatically reopen a closed ticket so nothing falls through the cracks.
When a connected platform has been consistently refusing requests, your integrations page now shows a warning so you know something needs attention — instead of silently failing in the background. Stores whose Shopify subscription is inactive are automatically paused to stop wasting credits, and resume on their own once Shopify confirms the shop is available again.
Before you leave for Facebook, the connection flow now tells you exactly what role you need on each Page and where to set it in Business Suite. The Accounts page shows what each connected account can do rather than internal status messages, and Instagram accounts are shown under their parent Page so the relationship is clear. Pages you cannot connect yet are grouped into a collapsible summary with the one step that fixes them.
When Otto uses tools to look something up before answering, it now sends a short orienting line first so you know it is working. Previously the screen could stay blank for over ten seconds while Otto gathered information.
You can now ask Otto to explain the Brain's current strategy, standing instructions, schedule, why it held off last cycle, and what is queued next. You can also tell Otto to add, remove, or leave a one-off note for the Brain, and those instructions take effect on every future cycle. Changes are logged so there is always a record of who asked.
Workspace owners can now enable a shared credit pool so every store in the workspace draws from a single balance instead of each store managing its own. Turn it on from Workspace Settings → Billing, and each store's billing page will show its usage against the shared pool total. New stores you add later join automatically — no extra setup needed.
Workspace owners can now enable a shared credit pool so every store in the workspace draws from a single balance instead of each store managing its own. New stores added later join automatically — no manual setup needed. Turn it on from Settings in your workspace, where you'll also see a full breakdown of how credits are being used across your stores.
When a teammate with view-only access signs in, a one-time banner now tells them upfront that they cannot make changes — before they run into a confusing blocked action. The banner is dismissible and only appears once per browser, so it informs without becoming background noise.
Workspace owners who have been granted access to the Store Launch Engine can now use it directly from their workspace — no special staff permissions required. If your workspace has been opted in, the Add Store entry point will appear in your workspace settings.
Your subscribed webhooks will now receive an `order.refunded` event whenever a refund is recorded on an order, whether it was processed by Magistry or applied directly in your Shopify admin. The event is delivered exactly once per refund and includes a `settled` field that tells you whether the payment gateway had settled the refund at the moment it was first seen.
The checks Magistry runs before answering your first chat message now happen at the same time instead of one after another. You should notice a shorter wait between sending a message and seeing Otto begin to reply.
The Otto mode you pick — fast, smart, or deep — is now remembered for each store separately, so switching to deep on one store won't quietly follow you to another. Deep mode now uses extended thinking for more thorough answers, and each mode has a firm answer-length limit so a long report can no longer be cut off mid-sentence.
Each API response includes an `x-request-id` header, and that same ID is recorded against the matching audit row. You can send your own ID in the request header so both sides share it without translation. The ID also appears in the response body of the diagnose endpoint, because its output is pasted into support conversations where a header is not visible.
When a customer contacts you about a refund, you can now look up that specific order and amount in the agent activity export — not just that a refund happened somewhere that day. The export is available as JSON or CSV from the audit page, and every export is itself recorded so you have a complete picture of who reviewed what.
You can now rename any AI conversation from its row menu, with the name editing in place so you keep your place in the list. Pasting multi-line text collapses it into a single name rather than cutting off at the first line. If a name cannot be saved you are told immediately and the field stays open so you can correct it.
Otto now offers three distinct modes — Fast, Smart, and Deep — each with genuinely different reasoning depth and time budgets. Fast gets you a quick answer with fewer tool rounds; Deep works harder and longer on complex questions. A mode picker in the chat input lets you choose before you send.
If you create a webhook subscription whose event-type filters only match types that have no active producer, the API now returns 201 with a warning listing the dormant types by name. You can still subscribe ahead of a producer shipping, but you will no longer receive silent confirmation that implies events are on their way.
AI conversations now get a meaningful name based on your first exchange instead of showing your raw question as the title. The name is set once and never overwrites anything you have chosen yourself. If a good name cannot be generated, the conversation keeps what it had.
When creating an API key you now choose an expiry — defaulting to 90 days — so keys no longer live forever by accident; 'Never' remains available for integrations that genuinely need it. Rotating a key now keeps the old key working for seven days alongside the new one, so you can redeploy at your own pace without an outage. The successor inherits the same scopes and settings exactly, and the rotation is blocked if the key is already revoked or mid-rotation.
You can now set an expiry date on headless API keys, restrict them to a list of allowed IP addresses or CIDR ranges, and rotate them with an overlap window so your integration stays live while you deploy a replacement. When a valid key is rejected, the error now tells you whether it expired or was blocked by an address rule — without leaking that information to unknown callers.
You can now copy any answer Otto gives you, ask the same question again if the answer missed the mark, and tell Otto whether a response was useful or not. Thumbs down opens an optional box where you can say what went wrong — your feedback goes somewhere it is read, not into a void.
When Otto produces a report, table, chart, or plan, it can now open it beside your conversation as an artifact rather than leaving it buried in the thread. Artifacts stick around after the chat ends, every revision is saved with a note explaining what changed, and you can copy or download the whole thing as Markdown at any time.
When you edit a grant — removing a store or dropping a scope — apps that connect via the Magistry MCP endpoint can now detect the change immediately from a version token in each tools response, and re-read only when something actually differs. Pagination is also live for credentials with read and write access, so every available tool is reliably returned across multiple pages.
When you ask Otto about your data, it can now respond with a chart — line, bar, area, pie, donut, scatter, and more — chosen to match what the data is showing. Charts include a title, currency-aware formatting, and a data table so you can always see the numbers behind the visual. Charts are saved to the conversation and will still be there when you come back.
AI clients like Claude and Cursor can now read each MCP tool's risk level before deciding whether to ask for confirmation. Write operations — including refunds and price changes — are always flagged as potentially destructive, so a client will prompt before acting rather than treating them the same as a read. Repeated calls to write tools are also marked as non-idempotent, meaning a second refund call is correctly treated as a second refund.
A new "Test connection" button on the Developers page runs four checks in the order they actually block a connection — live grant, exposed surface, credential validity, and recent activity — and tells you whose problem each step is. Any step that could not be evaluated is shown as unreadable rather than passed, so a green result means the connection genuinely works.
The importer now resolves your columns itself where the file can settle them, matches on the Shopify product id when the sheet carries one, and reads a cost per variant instead of broadcasting the first row's price across every colour. A sheet priced in a currency your shop does not sell in is recorded and held back rather than written to Shopify as if it were the shop's own.
An account Google has switched off is not a feed-quality problem, and it used to be graded with the same report as a healthy one. The compliance page now leads with the account's standing and a day counter, sequences the steps that cannot be reordered, and says plainly which requests Magistry can make for you and which only you can press.
Storefront presets are a lever the Brain can propose like any other: it snapshots the exact files first, the edit carries a revert time taken from your own promotion calendar, and a timer performs the revert. One change per store per day, so a promotion window yields a reading rather than a pile-up.
A 200 from Google means the request parsed, not that the conversion was accepted — acceptance, the per-event errors and the match rate arrive later. Every upload now leaves a receipt, an hourly sweep settles it against Google's own answer, and the tracking doctor reports rejected, unsettled and accepted-but-unmatched as three findings, because they have three different fixes.
A workspace can hold client stores with their own access requests, pooled credits and per-client reporting. It is opt-in per workspace and off by default: reaching another merchant's store is not a switch a workspace owner should be one click away from.
The launch engine takes a niche, researches a name with a free matching domain, provisions the accounts the store needs and works through the build as a sequence of steps you can watch and resume. Where a step ends at a person — accepting an invitation, adding a card — it says so instead of reporting itself finished.
A thumbs-down removes that reply from the examples the agent learns your voice from, immediately, rather than ranking it lower and leaving it in the corpus. Writing the reply that should have gone out instead teaches the desk directly; the correction is offered and never required.
The command palette (Cmd+K) now uses fuzzy search across many more page types and entity kinds, so you find what you need with fewer keystrokes. Results are grouped by category, and a curated zero-state shows your most useful pages when nothing is typed. Quick actions for dark mode, feedback, help, changelog, and sign-out are also available directly from the palette.
The Marketing Brain now runs on Claude Sonnet 5, Anthropic's latest model. You can expect sharper campaign proposals and more relevant recommendations when the Brain generates your marketing strategy.
You can now mark each log entry as a Refund (money back, item kept) or a Return (item shipped back), with carrier and tracking fields that appear only when relevant. A new All / Refunds / Returns filter and per-row badges make it easy to spot what came back. You can also log several products in a single record using the new Add Product rows, and the detail page shows a full per-product breakdown with totals.
You can now export four new entity types from the bulk configurator: Files (images, videos, and generic files with URLs and alt text), Menus (Online Store navigation flattened to one row per item), Companies (B2B company records with contact details), and Payouts (Shopify Payments history with date, status, and net amount). These are export-only sheets — import and push are not yet supported for these entities.
You can now export four additional entity types from the bulk export tool: store details, draft orders (with line-item rows and date/tag filtering), discount codes and automatic discounts, and metaobjects with their custom field columns. These are available under a new Store group in the export entity picker.
Bulk import and export are now accessible directly from the Products page and the import flow, so you no longer need to hunt through the sidebar. Every entry point opens the right tab automatically, and the same options are also available under Settings for quick access.
When exporting, you can now pick exactly which groups of columns to include for each entity — for example, export Products with just Basic and Inventory data, or Customers with Addresses but without Metafields. The Basic group is always included so your files stay re-importable. Exports with no selection made behave exactly as before.
You can now bulk-export Blog Posts (and import them back) and export your historical Orders as a spreadsheet. Exports can be downloaded as Excel or CSV, and you can filter Products and Orders by status, tag, vendor, date, and more before exporting. The bulk jobs page also shows clearer labels, relative timestamps, and automatically resumes tracking any jobs still in progress when you return to the page.
// direction
// stay current
// changelog list
// e-mails not sent yet · no tracking pixels
// see it run
Connect your store and choose the Brain's mode — in dry run every decision is logged with what it would have done, and nothing touches your store.
The Brain changes nothing until you choose a mode · Every action logged · Price, status and copy changes can be rolled back
We use cookies to run this site, to see which pages help, and to measure which ads bring people here. You choose per category; everything except the strictly necessary is off until you say otherwise. Cookie policy · Privacy
