FlowX.AI 5.12 is a SaaS release. New capabilities are available on managed (SaaS) deployments; self-hosted deployments will receive them with the next LTS release family.
Deployment models:
- SaaS Managed FlowX cloud deployments
- All Deployments Available in both self-hosted and SaaS
What changed since 5.11.0
What’s new?
- Integrations & Data
- AI Platform
- Platform
🔀 Multi-instance subworkflow node - Fan a workflow out over a collection
⚡ Event-triggered workflows - Start a workflow from an email or webhook
🗄️ Db2 relational data source - Connect to Db2 for i or Db2 LUW
🔻 Subworkflow failure exit - Send a failed mapping down its own branch
🔓 Skip TLS verification on a REST data source - Reach a trusted internal host with an untrusted certificate
🔔 Send Notification node - Send a templated notification straight from a workflow
⚡ Event-triggered workflows - Start a workflow from an email or webhook
🗄️ Db2 relational data source - Connect to Db2 for i or Db2 LUW
🔻 Subworkflow failure exit - Send a failed mapping down its own branch
🔓 Skip TLS verification on a REST data source - Reach a trusted internal host with an untrusted certificate
🔔 Send Notification node - Send a templated notification straight from a workflow
Multi-instance subworkflow node
Multi-instance subworkflow node
- Overview
- Key Features
Before 5.12.0: running an integration workflow once per item in a collection meant leaving the Integration Designer, wrapping the workflow in a BPMN process and fanning out with a multi-instance Call Activity. Now: the Subworkflow node fans out over an array on its own, in parallel.Turn on Parallel Multi-Instance under Execution on a Subworkflow node, pick the Input Array to iterate, and pick the Output Array that collects each run’s result. The node starts one subworkflow instance per element of the input array and gathers the results into the output array on the parent workflow.
Event-triggered workflows
Event-triggered workflows
- Overview
- Key Features
- Architecture
Before 5.12.0: an incoming email or webhook could start a process. Now: the same triggers can start an integration workflow directly.A workflow declares the trigger it listens for, and the platform keeps a subscription registry so a change to that trigger takes effect without rebuilding the whole app. Incoming email and webhook events are routed to the workflow that subscribes to them.
Bulk process-instance migration
Bulk process-instance migration
- Overview
- Key Features
Before 5.12.0: moving running process instances onto a new build was a one-at-a-time operation. Now: a filtered set of instances migrates in a single run.Select the target build, filter the instances to move, check the migration for problems before committing to it, and then run it. Progress and per-instance outcomes are available while it runs, and terminated instances are handled explicitly rather than failing the batch.
Documentation
See migrating process instances.
Runtime list filters and a Workflow Instances page
Runtime list filters and a Workflow Instances page
Finding a failed run no longer means scanning an unfiltered list. The Process Instances and UI Flow Sessions lists gain a multi-select Status filter and a start-date filter on their columns, so “every failed instance since last night” is two clicks.Workflow runs get more: a new Workflow Instances page under Runtime lists every run in the project, filterable by Status, by date, and by Origin: whether a run was started by a process, a UI Flow, a webhook, or an email trigger. Until now, a workflow’s runs were only visible from inside that workflow in the Integration Designer.For details, see the Workflow Instances documentation.
SaaS
Personal-data mask and restore nodes
Personal-data mask and restore nodes
- Overview
- Key Features
Before 5.12.0: personal-information handling was configured on the agent rather than placed in the flow. Now: masking and restoring personal data are workflow nodes you position explicitly, and you choose which key from the node input they apply to.Put a mask node before the step that must not see the raw values and a restore node after it. The workflow run log shows what each node did, so a run can be inspected without exposing the underlying data.
Documentation
See the Personal Information Guard documentation.
Db2 relational data source
Db2 relational data source
IBM Db2 joins the relational database data sources available in the Integration Designer. Because Db2 products use different drivers, ports and catalogs, the connection starts with a Db2 Product choice and the rest of the form follows from it. Db2 for i (IBM i) is the default and Db2 LUW is also supported. Db2 for z/OS appears in the list but is not implemented in this release, and selecting it fails with an explicit error rather than falling back to another driver.The port is prefilled from the product and whether SSL is on. Db2 for i adds a naming convention, an optional library and library list, and a switch for returning character data that the host reports as binary. On Db2 for i, connections also open read-only by default, so turn that off if your queries need to write, and the transaction isolation level defaults to none, because a table without a journal cannot take part in a transaction. Isolation is validated when the data source is saved rather than when a query first runs.For the full configuration reference, see the Db2 Database documentation.
SaaS
Subworkflow failure exit
Subworkflow failure exit
A Subworkflow node now has a second exit. Connect the red one and a failure takes its own branch instead of stopping the parent workflow. It is opt-in per node, so a node without that branch fails and blocks exactly as before and nothing changes for existing workflows.Only a mapping failure routes, which is the kind a workflow can be built to handle: the input mapper failing, the input not satisfying what the subworkflow expects, or the output mapper failing on merge-back. An unset subworkflow, a missing response key, no mapping at all, and the node and depth limits all still stop the run. The branch reads a fixed
subWorkflowError key that says which mapping failed and, when a child did start, which run it was.For the full configuration reference, see Routing a subworkflow failure.
SaaSSkip TLS verification on a REST data source
Skip TLS verification on a REST data source
A REST data source can now be told to stop checking the certificate of the host it calls, for the cases where an internal endpoint presents a self-signed or mismatched certificate and the alternative is an infrastructure workaround. It is off by default and set per data source, so turning it on for one connector leaves every other call verified.The Designer warns while it is on, because the call can no longer tell the intended host from anything intercepting it. One consequence is worth knowing before you use it: a truststore configured under Certificates is ignored while verification is skipped, since skipping replaces it rather than adding to it.For the full configuration reference, see TLS verification on a REST data source.
SaaS
Send Notification node
Send Notification node
Workflows can now send notifications themselves. A Send Notification node, under Tools in the workflow node palette, takes one of your project’s notification templates, fills its variables from workflow data, and sends it, without routing through a process just to reach the Notifications plugin.Address it with To, Cc and Bcc, as fixed addresses or
${} keys. Email options are all overridable per node: subject, sender name and email, attachments by file path or ${} key, and custom headers. The node reports the outcome under its response key, including recipient counts and the error when sending fails, so the workflow can branch on it.For the full configuration reference, see the Send Notification node documentation.
SaaSReusable functions as agent tools
Reusable functions as agent tools
An AI node’s tools can now include a Function, alongside built-in web search, an MCP server, and a workflow. Point an agent at a reusable function when the work is deterministic and you would rather it call your code than reason its way there.A function has to be typed to qualify: it needs both an input and an output slice, and functions without them do not appear in the picker. Give it a description as well, since that is what tells the agent when to reach for it.For the full configuration reference, see the Custom Agent node documentation.
SaaS
Chat response feedback
Chat response feedback
Users can rate an agent’s answer in the chat component, and follow the rating with a reason. The ratings are recorded in audit, so the quality of a deployed agent becomes something you can look at rather than guess at.The feedback controls and the reason chips are themable, with their own selected and hover states, so they can be styled to match the rest of the interface. The SDKs expose the same feedback state to custom front ends.For the full configuration reference, see Rating an answer on the chat component.
SaaS
Custom evaluators
Custom evaluators
The evaluator catalogue is no longer fixed. When none of the built-in rubrics measures what you care about, you can write your own evaluator, give it the scoring instructions the judge LLM should follow, and it appears in the catalogue under a Custom group alongside the built-in ones.Choose Numeric scoring for a score between 0 and 1 with a pass threshold, or Binary for a straight pass or fail. You can start from an existing evaluator to adapt a built-in rubric rather than writing a prompt from scratch. Custom evaluators belong to the workflow you create them in, and unlike the built-in ones they can be edited and deleted.For the full configuration reference, see the Evaluations documentation.
SaaS
Deep Research node
Deep Research node
A new AI node researches a question across the open web and files a cited report into a Knowledge Base. It plans the work, sends sub-researchers out to search and read pages, grades what they bring back, and keeps going until its budget is spent or it has enough to report.Unlike a Custom Agent, you do not write the method: the research prompt is built into the node, so you describe the subject and bind workflow data into it. Depth is a Quick, Standard or Deep preset, or Custom if you want to set the number of passes and tool calls yourself. You can restrict which domains the research may read, with an allow list, a deny list, or both.The result always lands in a Knowledge Base as text, so retrieval nodes and other agents can reach it later. Rendering it as a document is optional.For the full configuration reference, see the Deep Research node documentation.
SaaS
Built-in agent tools by category
Built-in agent tools by category
A Custom Agent’s built-in tools are now a catalog you pick from by category: Math tools, List tools, Date tools, Text tools, and Web search. Each category row shows exactly which tools it turns on, and the agent runs deterministic code for that work instead of estimating it in the model.The catalog holds twenty tools, fifteen of them new: a full date family (business days, date arithmetic, formatting, parsing, timezone conversion), text helpers (regex match and extract, fuzzy match), unit conversion, and list aggregation, joining the original calculator, compare, sort, and filter. Web search is part of the same catalog rather than a separate toggle.Selection stays opt-in and per node. Agents saved with the old toggles keep working without an edit.For details, see the Custom Agent node documentation.
SaaS
AI Designer builds UI Flows
AI Designer builds UI Flows
The AI Designer now works at the level of a whole UI Flow. Describe the journey and it generates a multi-page flow: structure first, pages filled in parallel, navigation wired between them. Ask for a change and it edits the affected pages instead of regenerating everything.Proposals appear on the UI Flow canvas as a draft for you to review, the editor stays locked while one is pending, and any data-model changes the flow needs are presented for approval on every turn, the same way single-screen generation already works.For details, see the AI Designer documentation.
SaaS
Check a version for issues
Check a version for issues
The Designer can now validate the resources in a project version and collect what it finds in one Issues panel at the bottom of the screen. Findings are marked where they occur, on resource lists, page titles and workflow canvas nodes, and selecting a marker opens the panel on that finding.Checks run when you ask for one, and the panel says when it last ran so a stale result is obvious. You can scope the list to the whole version, the resource you are in, or the node you have selected, filter it by severity or text, and group it by resource or severity. Opening Commit Changes to Version runs a check on its own and shows each changed resource’s state beside it, as a review step rather than a gate. Integration workflows are the resource type validated in this first iteration.For the full configuration reference, see the Checking a version for issues documentation.
SaaS
Branch management
Branch management
Project branches gained a management layer. The merge modal now offers every branch you can merge into, not just the current branch’s parents, so a fix made anywhere can land anywhere it is needed.Branches you are done with no longer linger: Archive Branch retires one (with a required reason, shown later in its info panel), Unarchive Branch brings it back, and Delete Branch removes one permanently, guarded so it refuses while child branches, an uncommitted draft, or a running merge depend on it, and asks you to type the branch name when unmerged work would be lost. Deleting has its own permission.Branch and version info panels round it out: a version merged from another branch says where it came from, including whether that branch has since been archived.For details, see Managing branches.
SaaS
Process canvas editing
Process canvas editing
The BPMN canvas gains a set of direct editing improvements. A New Node panel and
drop-anywhere connecting make it quicker to add a node and wire it up, connecting to a nearby
node no longer means aiming precisely at a handle, and sequence flows can be reshaped by
dragging their control points. Swimlanes now resize from any of their four edges. The
process-instance viewer picks up the same canvas.The canvas is deliberately unchanged to look at, so existing diagrams, habits and screenshots
still hold.For the full configuration reference, see Faster ways to build on the Process Designer canvas.
SaaS
Reusable UI Templates in UI Flows
Reusable UI Templates in UI Flows
A Reusable UI Template can now be built for UI Flow pages as well as for the User Tasks in a process. Each template declares where it is used when you create it, and the UI Flow editor can drop templates onto a page the same way the Process editor always could.The host type also decides which actions a template’s event handlers offer, so a template built for UI Flows gets navigation and workflow actions while one built for processes gets node actions. Because of that the choice is fixed once the template exists, and templates you created before this release count as Process templates, so nothing you already built changes.For the full configuration reference, see Where a template can be used in the Reusable UI Templates documentation.
SaaS