October 2026

Changes since the July 2026 release, across the platform, connectors, agents and the admin experience.

• New Connectors & Integrations
  • Invoicing over the SAP Ariba Network
• Mapping & Data Sync
• Performance
• Warehouse & Inventory Data Model
• Configuration Builder Experience
• AI & Assistant Data Access
• Search & Indexing
• Agents & On-Premise
• Reliability & Monitoring
• Documentation
• Upgrade Notes

New Connectors & Integrations

New connectors: Epicor Kinetic (manufacturing ERP), Epicor Prophet 21 (distribution ERP), Jira (issue tracking), Ordoro (order & inventory management), and Finexio (payments). Sage Intacct now runs on that vendor's REST API.

• Failures that used to look like successes are now caught — Some systems accept a write with an HTTP 200 and report the rejection inside the response body. Central used to record those as successful syncs. It now inspects the response and fails the sync, so a record the other system actually refused no longer shows as delivered.
• You can see why a system said no — When a connected system rejects a request, you get its own explanation rather than a generic wrapper. More than half the connector catalogue previously surfaced no error text at all; those now report a usable message, and Sage Intacct and Rent Manager failures in particular are no longer blank.
• More systems can notify you of changes — Webhook support was added for Payload and Loopio and enabled for Paycor, so those systems can push changes to you instead of waiting for the next scheduled poll.
• Large reads are more dependable — BILL now pages through long result sets correctly, Wherefour retries instead of failing when it asks you to come back later, and QuickBooks Time accepts the dates Central sends.
• Fixes — Cincom CPQ connects again, Active Directory / LDAP endpoints can be created, and picking a system that comes in several flavours (SAP Ariba, Rental Result) no longer errors.

Invoicing over the SAP Ariba Network

Central can now send cXML invoices to the Ariba Network, including the PDF the buyer expects attached to them.

• Attach the invoice document — Author attachments directly in the endpoint configuration, and Central sends the invoice and its PDF together as a single request.
• Match invoices to their purchase order — Invoices link back to the originating PO, including when the buyer's document arrives without a payload identifier.
• Control the document header — Set the XML declaration and DOCTYPE per endpoint where a trading partner requires its own.

Mapping & Data Sync

• Match incoming records on more than their keys — When an inbound message carries no key Central recognises, it can now look the record up by properties you nominate — an email address, an account code, a reference number. The key is attached to the record already there instead of a duplicate being created. More than one match is reported as an error rather than silently taking the first.
• Put a stored setting or secret into a request — A new Connector Setting map source places a connector setting straight into a request body, on reads as well as writes, with secrets kept out of logs. Useful for systems that expect a shared secret in every call rather than in a header.
• Decline records you can't place — A new Unresolved Record Handling option lets an endpoint refuse to create a record whose reference it can't resolve, so a half-formed orphan isn't created while the real record is still on its way.
• Page the way each endpoint needs to — Choose a paging approach and page size per endpoint instead of inheriting the connector's, from a dedicated Paging tab.
• Collapse a collection into one value — Render a list of values coming from a connector into a single field.
• Apply a filter only to certain operations — Scope a subscriber filter so it applies to new records but not updates, or the reverse.
• Resolve references by Central's own ID — Useful where the two systems share no common identifier.
• Borrow a value from the parent record — Map From Parent now works on looping-list maps, with the right parent fields offered in the picker.
• Date transforms are selectable again — The date-adjustment and date-only transforms were missing from the endpoint transform picker, and date-adjustment rejected text dates. Both work.
• Fewer silent surprises — An empty source value no longer produces a blank downstream record; a property that can't be resolved fails loudly instead of being quietly set aside; an empty XML attribute writes as empty instead of failing the whole document; and a map naming its own system is no longer dropped from key replication.
• Keys arrive more reliably — Keys captured on child records now replicate, a key is only reported when its value actually changed, and a record carrying duplicate keys for one system returns a clear error instead of a retried server failure.

Performance

• Large reads are dramatically faster — A rebuilt connector read path cut a representative large read from 22 seconds to under one, and its memory use by about 94%. Big paged pulls and wide line-item sets finish faster and are far less likely to time out.
• Faster cross-entity questions — Queries that filter one entity by a related one now run as a single operation instead of one query per row; a report that took around twenty seconds returns in about one.

Warehouse & Inventory Data Model

• New warehouse task types — Receiving Task and Putaway Task join Pick, Pack and Wave as types of Warehouse Task, and two new entity types arrive alongside them: Warehouse Task Line and Inventory Balance. Inventory Change Event and its lines are also now modelled.
• Richer sales order lines — Sales order line items carry cost, margin and fulfillment quantity.
• Renamed for consistency — The Pick, Pack and Wave task types introduced in v5.15 are now Pick Task, Pack Task and Wave Task, and the Wave Item entity type is now Warehouse Task Line. See Upgrade Notes.

Configuration Builder Experience

• Start from a worked example — Shipped endpoint templates for Paycor, Office 365, Deltek Vantagepoint and QuickBooks Time give you a configured starting point instead of a blank endpoint. Templates can also carry universal filters.
• Run a subscriber over just what you need — Trigger a run with filters applied and variables overridden, instead of re-running everything to re-send a handful of records.
• Reorder mappings by dragging — Drag to reorder maps within a list, and to move them into child groups.
• Send one record to several destinations — Where a record carries a list of destinations, configure Central to deliver it to each of them.
• Author projections from the parent — Build a projection starting from its parent record, with its type field filled in for you.
• Export your data model — Produce a document of the common model you're working against, including Excel, for anyone documenting the integration.
• Settings organised by direction — The Endpoint Settings tab is grouped into inbound and outbound, so you only see what applies.
• Templates that finish the job — A data provider added from a template now creates and binds its data model instead of arriving unbound.
• Caught earlier — Duplicate endpoint IDs and names are flagged within and across hubs, and a read-shaped endpoint that isn't typed as a data provider is called out.
• Fixes — Delete spinners stop when a delete fails, monitor date-range grids reload with the range you actually picked rather than the previous one, and Maintain Index shows schemas on deprecated entity types again.

AI & Assistant Data Access

• Ask for a total without reading every row — Totals, averages and counts can now be computed over a saved query's results in a single answer. A question like "total hours in August per person" over hundreds of entries is answered directly, where before the underlying rows wouldn't fit in one response.
• Scope a saved query to whoever is asking — A saved query can filter to the signed-in user, so the same question returns each person's own data.
• Filters return what you'd expect — Comparisons against numbers, dates and true/false values work rather than erroring, and date ranges return real results instead of silently returning nothing.
• Date comparisons across related records are correct — Same-day records are no longer counted as late.
• The assistant describes your data accurately — Field lists are complete, a repeated property name no longer takes out a whole data model, and related records are named so two different relationships can't be confused for one.
• Use Central from ChatGPT — Central can be added as a connector in ChatGPT, with sign-in through the standard authorization prompt.

Search & Indexing

• Searches cover everything they should — Properties arriving through webhooks and inner actions are now included when the search index is built; they were previously skipped, so they could not be found or matched on.
• Backfills cover the whole entity — A publisher nested inside another publisher is now included, so a backfill no longer rebuilds only part of an entity type.

Agents & On-Premise

• Settings changes stick on self-update — Updating an agent can now change which API it talks to; that setting previously reverted.
• Update an agent from the command line — central agent update triggers a self-update without the manual steps.
• A smoother Windows install — The installer finds your existing agent service rather than asking you to name it, completes an upgrade instead of hanging, and tells you what's wrong when it can't proceed.

Reliability & Monitoring

• See exactly what a publisher read — Publisher runs record each source file consumed and the date window applied, and that window is shown in the UI.
• No files missed or read twice at a window edge — A file landing exactly on the boundary of a read window is now picked up once.
• Line-item order is preserved — Line items come back in the order they arrived, through storage and reload.
• Fewer syncs lost without explanation — Simultaneous updates to the same record no longer conflict, a record with an empty child list no longer stops the sync, and a connector error that isn't JSON no longer destroys the record of what happened.
• Clearer sign-in failures — An ambiguous email address is refused with an explanation instead of a server error, sign-in no longer fails intermittently, and errors during sign-in are shown rather than swallowed.

Documentation

• Looping Lists & Custom Lists — A new guide with worked examples from source data to model.
• Value Formatting appendix and a completed transforms reference table.
• Inner actions and their filtering are now documented.
• Central Support page added to the guide.

Upgrade Notes

Warehouse task renames. If you configured mappings against the Pick, Pack or Wave task types introduced in v5.15, they are now sub-types of the WarehouseTask entity type, named PickTask, PackTask and WaveTask. Re-select the entity type as WarehouseTask and pick the matching sub-type. Separately, the WaveItem entity type is now WarehouseTaskLine.

The associated key names changed with them — PickId → PickTaskId, WaveId → WaveTaskId — and PackTaskId, ReceivingTaskId, ReceivingTaskLineId and PutawayTaskId are new. There is no automatic alias from the old names. Contact Central Support if you'd like help confirming whether this applies to you.

Search re-index recommended. Search indexes built before this release omitted properties arriving through webhooks and inner actions, and backfills for nested publishers covered only part of their entity type. If you search or match on those properties, run a backfill after upgrading.