07/10/26 Release Notes (core)
Release date: 7th October 2026
New Features
Custom Scripting - new UI and AI experience
We've rebuilt the custom scripting experience, making scripts easier to create, safer to change and simpler to keep track of.
Typed scripts - choose the type of script you're creating (custom, pre/post authentication, pre/post request or pagination) and the editor shows only the inputs and variables relevant to that type.
Save without deploying - saving a script no longer affects live flows. Save as often as you like, then deploy when you're ready. Test deploys let you try a change before committing to it, and deploys that hit a build error are retried automatically.
Version history and diffs - the last ten versions of each script are retained, and a built-in diff viewer shows exactly what changed between any two.
See all uses - view every flow, connector endpoint and mapping that references a script before you change it.
Type-aware testing - the testing sidebar adapts to the script type, so you're always testing with the right inputs.
Build logs and audit trail - build and compile output is surfaced in the editor, and every edit and deploy is recorded in the audit log.
Redesigned editor - a refreshed scripts page with full-screen editing.
API Keys - Roles & Permissions
API keys can now be given specific roles, so a key only has the permissions it needs.
Previously, every API key was created with manager-level access. When you create a key you now choose an API key type (standard or MCP) and assign one or more roles - either built-in Patchworks roles, or custom roles defined in the Role Manager. The key can only perform actions permitted by its roles, and the API keys list now shows the scope and roles for every key at a glance.
Existing keys are unaffected - they've been assigned roles matching their previous access, so nothing changes for integrations already using them.
Platform Agents - Migration Agent
The Migration agent is now a dedicated Platform Agent, available directly from the AI Assistant and the Platform Agents page. Previously it was reached through the Implementation Agent.
Share what you have today - exports, specifications, mapping documents or a description of the current integration - and the agent audits it, proposes the equivalent Patchworks connectors and process flows, and works with the Implementation Agent to build them. You confirm the proposal before anything is created.
Process Flow Level Concurrency Control
You can now cap how many instances of a single process flow run at the same time. A new maximum concurrent runs field in process flow settings (under advanced settings) lets you set a per-flow limit within your account's concurrent process flows allowance. When the limit is reached, further run jobs wait in your queue rather than running in parallel, so scheduled flows no longer stack up after a backlog and parent flows that spawn a child run per record won't overwhelm downstream systems. The setting is optional - flows without a value behave exactly as before.
Set the limit to 1 to guarantee that only one instance of a flow runs at any time.
Improvements
AI Studio - AWS Bedrock as Primary Model Provider
AI Studio's frontier models are now served through AWS Bedrock, inside the Patchworks AWS environment, rather than via a third-party API. This brings AI Studio under the same infrastructure, security controls and per-tenant isolation as the rest of the platform, and requests processed through Bedrock are not used to train the underlying models.
Bedrock is also the first step in the provider failover introduced last month: if Bedrock is unavailable, AI Studio falls back to the Anthropic API, and then to the OpenAI API, so the service stays operational throughout. Nothing changes in how you use the agents. For more detail, see the new Models page.
Webhooks - Request Headers Available as Meta Fields
Flows triggered by a webhook can now read the inbound HTTP headers. Headers are stored on the payload meta as request_headers and can be referenced anywhere [[meta.*]] variables are accepted - maps, filters, Set Variables, Notify shapes, scripts and connector fields - using the lowercase header name, for example [[meta.request_headers.x-shopify-topic]]. This gives you access to values such as a webhook topic, signature or shop domain that aren't in the request body.
Header names must be referenced in lowercase. Cookie and Authorization headers are never stored, and unusually large header sets are dropped so the webhook still triggers.
Notify Shape - Environment Name in Emails
Emails sent by the Notify shape now include the virtual environment name when the flow ran in one. The subject takes the form [Environment] Warning alert for {flow name} (#{run id}) and the body includes an Environment row, so you can tell at a glance which environment an alert came from. Emails from runs without a virtual environment are unchanged.
Company Switcher - Recent Companies
For users who manage multiple linked companies, the company dropdown now includes a Recent companies section listing the companies you've most recently switched between, so you can jump back without searching the full list. The list is kept in your browser, so it's specific to the device you're using.
Health Monitor Agent - Run ID Included in Diagnose with AI
When you open Diagnose with AI from the Flow Runs list or a run log page, or Debug with AI from an error row in the step logs, the pre-filled question now includes the run ID of the run you came from. The question is still editable and nothing is sent until you press send.
Bug Fixes
Fix: Try/Catch Not Catching Failures When Starting a Flow Mid-way
Fixed an issue where starting a run from a step inside a Try/Catch block (using start from here) meant connector failures were not caught - the run failed and the remaining records were not processed. Failures inside the Try block are now routed to the Catch branch regardless of where the run was started, and the connector continues through the remaining records.
Fix: Flows Could Be Deployed Beyond the Virtual Environment Limit
Fixed an issue where a flow could be deployed to more virtual environments than your subscription allows. The limit applies to environments that have at least one flow deployed to them - creating environments and publishing their configuration versions is not limited. Deploying a flow to an additional environment beyond your limit is now blocked with a clear message (in the UI and via the API) and no orphaned flow version is created, while deploying to environments already in use continues to work. The virtual environments page now shows Active Virtual Environments - n / n so the count is clear.
Note that an environment counts towards the limit while it holds any deployment, enabled or disabled. To free a slot, remove the deployments from an environment you no longer use.
Fix: Environment Variables Empty in Map Scripts
Fixed an issue where virtual environment variables were not populated when referenced in a script used within a Map shape, even though the same script worked as a standalone Script shape. Environment variables now resolve consistently in both.
Fix: Database Connector Failing on Datetime Values
Resolved an issue where database queries containing a time value - for example WHERE created_at > "2026-01-01 00:00:00" - failed with SQLSTATE[HY093]: Invalid parameter number, because the :00 in the time was being treated as a query parameter. Datetime values in queries now run correctly.
Fix: List-type Optional URL Parameters Not Displaying
Fixed an issue where the dropdown for list-type optional URL parameters did not render in the connector shape, so an option could not be selected. The list now displays as expected.
Fix: Connector Export Failing for Some Companies
Fixed a database error that prevented some companies from exporting a connector. As part of this change, connector exports no longer include endpoint field tags used by the Track Data shape; exports created before this release that still contain them can be imported as before.
Fix: Event Connector Form Accepting Empty Required Fields
Fixed an issue where the event connector form allowed Test and Save with required fields left empty - such as a Basic auth username or password, or certificates for TLS/x.509 - resulting in a broker connection error rather than a field error. Required fields across the details, authentication and queues sections are now validated first, and previously saved passwords and certificates still count as filled.
For questions about this release, please contact the Patchworks support team or refer to the product documentation.
Last updated
Was this helpful?