FlowX.AI 5.9.x to 6.0.0 changelog

​

New

• Developer Agentic Harness: a plan can split the work into stories, delivered one at a time, each with how to test it. Comment mode lets you select a component in the running app and ask for a change, and conversations follow you across the Designer. See Developer Agentic Harness.
• Integration Designer: add schedules to a workflow’s Start node to run it once on a date or repeatedly on an ISO or cron cycle. A schedule fires only after you activate it in Runtime → Runtime Settings → Schedules. See How do I run a workflow on a schedule?.
• Integration Designer: the Generate Tabular File node turns an array of rows into a CSV or XLSX file. See Generate Tabular File node.
• FlowX Database: an index can enforce unique values on one attribute or a compound key, and existing duplicates are reported when you run Apply Indexes. You can also delete a document from the collection view. See How do I make an index unique?.
• Knowledge Base: Disable Entry takes an entry out of retrieval without removing it, Activate Entry puts it back, and an entry can be permanently deleted. See Disable, re-activate, or permanently delete an entry.
• Resources: Documentation in a resource’s menu holds a short description and longer Markdown documentation. Media Library assets get a description, used as the image’s alt text. See How do I document a resource?.
• Substitution tags: write ${tag:key} anywhere in a component’s text, next to static text and process data. The upgrade rewrites whole-value @@key references to ${tag:key}, and @@key keeps working as a whole value. See How do I reference a substitution tag in UI content?.

​

Changed

• Designer header: the header holds the project name, the Config / Runtime switch, the Build with AI and Comment toggles, and Run project. Branching, the current version, and Commit Changes move to a version block in the footer. See Header and version block.
• AI nodes: every AI node except Deep Research retries up to 10 times, 30 seconds apart, when its AI service or provider can’t be reached. Turn on Allow Retry to set your own count, interval, and backoff. See AI node retries.
• Chat-driven workflows: the chat message is no longer appended to an AI node’s prompt. Reference ${userMessage} in the node’s Context. See Context.
• Evaluations: each workflow has its own test cases, grouped by tags, with equals and equals as JSON asserts and LLM evaluators. The upgrade converts datasets to test cases with no checks, so add an assert or evaluator to each one before you run it. See Datasets from earlier versions.
• Certificates: the Certificates panel on REST and File Storage data sources has two parts: Trust (System CAs, Custom CA, or None (insecure)) and Identity for mutual TLS. Existing settings are converted automatically. See Certificates.
• Knowledge Base: turning Containing Personal Information on or off re-embeds the existing content with the new model. See Can I change personal information after content is ingested?.
• Issues Console: a check also covers the version itself: entry points, theme, library builds, and unused dependencies. See Which version settings does a check cover?.
• Notification center: covers version imports and build exports and imports, alongside version exports and process instance migrations. See What sends a notification.
• Configuration parameters: after you save a sensitive value, the Designer shows ******** in its place. The real value is still used at runtime. See What happens after you save a sensitive value?.
• Script engine: scripts on GraalVM stop at the same time limit as the native engine: 5,000 ms for JavaScript and 10,000 ms for Python by default. See Script size and execution limits.
• Canvas: Auto-arrange on the workflow canvas and Arrange on the process canvas show a layout preview before anything moves. See How do I auto-arrange the workflow canvas?.
• Timer events: timer nodes use the schedule editor. Dates and times are picked in local time and saved in UTC. See Which entry modes can I use?.
• Versioning: branch names can be up to 40 characters, up from 16. See Creating a new branch.
• Datepicker: Allow manual input lets users type a date in the pattern of the project’s default locale. Web only.
• Data mapping: typing a new attribute path, such as client.address.street, creates the data types it names when you save the mapping.

​

Removed

• AI agents: the Command Center, the standalone Analyst, Architect, Assistant, Designer, and Developer agents, and the AI buttons inside the editors are removed. Open the agent rail with ⌘K (Ctrl+K) or the sparkle toggle in the header. See Developer Agentic Harness.
• Evaluations: the Code evaluators are removed. Use equals in place of Exact match and equals as JSON in place of JSON match. Levenshtein has no replacement. See Evaluations.
• Export and import: exports from 5.1.x no longer import, and 5.1.x is no longer an export target. See Exports from 5.1.x.

​

Deprecated

• GraalVM script engine: deprecated. FlowX.AI 6.1.x is the last release family that supports it. Scripts on GraalVM keep working in 6.0.0; move them to the native engine. See Move scripts off the GraalVM engine.

​

New

• Developer Agentic Harness: an AI agent in the Designer builds and changes apps from a plain-language description. It surveys the app, drafts a plan for you to confirm, and runs the change on its own branch; merging and serving a new version stay with you. See Developer Agentic Harness.
• Notification center: a bell in the Designer opens a 30-day notification history with read state, search, and filters. See Notification center.
• UI Designer: copy UI components and paste them into another UI Flow, app, or context. The paste names anything it could not place, and pasted components get unique names.
• Versioning: committing or merging a version takes an optional description of up to 2000 characters, and a commit details view shows what a version contains. See Commit descriptions and details.
• Runtime: an operator with the corrective-actions role moves the token of a running process instance to another node. See Move tokens.
• Document classifiers: a classifier’s corpus can be exported and imported between environments, and viewed read-only in the Data tab. See What travels between environments.
• Integration Designer: the File Encryption node encrypts or decrypts a file with a password inside a workflow. The file path and password take expressions.
• UI Designer: a start workflow trigger takes a condition, so the workflow starts only when the condition holds.
• FlowX Database: the Intersects operator matches documents whose array field shares at least one value with a given list. See Intersects on array fields.
• Personal Information Guard: AI nodes with the input scan on log the request as the model received it, with placeholders in place of the detected values. See Request after PII anonymization.

​

Changed

• Access management: the Workspace corrective actions editor role replaces the Workspace operations editor role. Users and groups with the old role move to the new one automatically. See Workspace corrective actions editor permission matrix.
• Export and import: version exports bundle dependencies the same way build exports do, and the export dialog preselects the target platform version. See Version exports that bundle dependencies.
• Integration Designer: the Tabular Data Extraction node becomes the Data Parse node and also parses XML and JSON. See Data Parse: XML and JSON support.
• Knowledge Base: embedding jobs are scheduled fairly by default, so a large ingestion no longer delays small jobs queued behind it.
• File sources: Document and Image components share one source configuration. In process user tasks, the Document Plugin Path source takes a Documents plugin downloadPath directly. See Selecting the file to download.
• Task management: the task list searches, filters, and sorts on the translated labels it displays. See Search and sort on translated labels.
• Substitution tags: custom tags can no longer start with sys_, which is reserved for system tags. See Substitution tags.
• Elasticsearch: new process instance and audit log indices default to auto_expand_replicas 0-1, which overrides a fixed replica count. See Elasticsearch replica auto-expansion.
• Processes: the process name moves from Settings into a Details modal.
• User management: inviting many users at once has a reworked flow, and a sign-up link takes new users directly to the right solution.

​

New

• Integration Designer: turn on Parallel Multi-Instance on a Subworkflow node to start one subworkflow per element of an input array and collect the results in an output array. See Parallel multi-instance.
• Integration Designer: email and webhook triggers can start an integration workflow, not only a process. See External triggers.
• Integration Designer: a Subworkflow node can route a mapping failure to its own branch instead of stopping the parent workflow. It is opt-in per node. See Routing a subworkflow failure.
• Integration Designer: the Send Notification node fills a project notification template from workflow data and sends it. See Send Notification node.
• Db2 data source: IBM Db2 joins the relational data sources, with Db2 for i as the default and Db2 LUW also supported. Db2 for z/OS is listed but not implemented, and selecting it fails with an error. See Db2 database.
• Runtime: bulk migration moves a filtered set of process instances to a new build in one run, with a check before it runs. See Migrate process instances.
• Runtime: a Workflow Instances page lists every workflow run in the project, filterable by status, date, and origin. The Process Instances and UI Flow Sessions lists gain status and start-date filters. See Workflow instances.
• Personal Information Guard: mask and restore nodes place personal-data handling as explicit workflow steps. See Masking and restoring across several nodes.
• Deep Research node: researches a question across the open web and files a cited report into a Knowledge Base. Depth is Quick, Standard, Deep, or Custom, and domains can be restricted with an allow or deny list. See Deep Research node.
• Custom Agent: a reusable function with input and output slices can be an agent tool. Built-in tools are a catalog of twenty tools in five categories: Math, List, Date, Text, and Web search. See Tools tab.
• Chat component: users rate an agent’s answer and add a reason, and the ratings are recorded in audit. See Rating an answer.
• Evaluations: custom evaluators score with Numeric or Binary rules and belong to the workflow you create them in. See Custom evaluators.
• Issues panel: the Designer validates the resources in a project version and collects the findings in one Issues panel. See Checking for issues.
• Versioning: merge into any branch you can merge into, and archive, unarchive, or delete branches. See Managing branches.
• Process canvas: a New Node panel, drop-anywhere connecting, draggable control points on sequence flows, and swimlanes that resize from all four edges. See Faster ways to build.
• Reusable UI templates: a template can be built for UI Flow pages as well as process user tasks. Templates created earlier count as Process templates. See Where a template can be used.

​

New

• Custom Agent: attach FlowX workflows as tools, each with a hint that tells the agent when to call it. See How workflow tools run.
• Custom Agent: agentic knowledge base search breaks a question into sub-queries and retries with reworded queries when the first pass finds nothing. Show Reasoning records the model’s reasoning summary in the logs. See Agentic knowledge base search.
• Knowledge Base: a Knowledge Base with Containing Personal Information on can use a local embedding model, set up as a CUSTOM provider in the organization AI settings. See Containing personal information.
• Knowledge Base: the Update Knowledge Base node can delete entries that match a metadata filter, and its FAIL branch reports what failed and how many chunks were rolled back. See Delete entries.
• Document classifiers: define classes, upload training data, and train or import a model, then route each document page with the Document Classification node. See Document classifiers.
• Reusable functions: Script and Condition (Fork) nodes in the Integration Designer can call reusable functions. See Reusable functions.
• Integration Designer: the Subworkflow node can map input per start node and output per end node instead of raw JSON. See Subworkflow node.
• Data model: attributes support validators such as Required, Min/Max Value, Min/Max Length, Pattern, and Email Format, wherever the data model is used. See Attribute validations.
• Templates: document and notification templates use the structured data model, with typed attributes and an Input Parameters tab. See Managing notification templates.
• Chat component: the chat button shows an unread-message badge, the panel searches conversation history, and the input can be disabled by a condition. See Chat.
• UI Flows: a Navigate To action can require selected forms to pass validation before it navigates. See Navigate To action.
• UI Flows: with Allow multiple editors off, a running session is restricted to its owner. New flows default to single-owner; existing and imported flows keep the shared behavior. See Session owner.
• Process Renderer: a process-end trigger runs a configured action when the embedded process ends. See Process end handling.

​

Changed

• Runtime authorization: runtime API requests are authorized against the roles defined in the running build, and requests without a matching build role are denied. Enforcement is on by default.
• Chat-driven workflows: a conversational workflow must send at least one chat reply, or the run fails with an error. See Chat-driven workflows.
• Custom Agent: the Instructions field is limited to 10,000 characters. See Custom Agent node.
• Response caching: POST REST endpoints, file-storage LIST and DOWNLOAD, and Mongo FIND, FIND_ONE, and AGGREGATE operations are cacheable. Mutating operations are never cached. See REST endpoint caching.
• Email triggers: Email Trigger and Microsoft Outlook data sources filter attachments by file extension. Existing category selections are migrated automatically. See Attachment handling.
• Audit: UI Flow runtime sessions are audited in more detail, including workflow runs, chat messages, and conversation reads. See Audit.
• Export and import: exporting to an older platform version removes UI-template features the target cannot read. See Cross-version export and import.
• UI Flows: an expired session shows a session-expired screen with a start-fresh option.
• Renderers: enumeration lookups are batched into a single request.

​

New

• PostgreSQL data source: connect a workflow to a PostgreSQL database and run SQL query operations from workflow nodes. See PostgreSQL database.
• SOAP data source: import a WSDL from a URL or file, and call each operation from workflows with a SOAP Operation node. See SOAP System.
• FlowX Database: MongoDB aggregation pipelines group, join, and reshape documents beyond a single query. See Aggregation pipelines.
• Integration Designer: the Tabular Data Extraction node reads a spreadsheet or CSV file into structured JSON, one object per row. See Tabular Data Extraction node.
• Integration Designer: the Template Filling node generates a PDF or DOCX document from a document template filled with workflow data. See Template Filling node.
• Integration Designer: the Combine Files node merges PDFs in a given order, and the Split File node splits a PDF by page ranges or per page. See Combine Files node and Split File node.
• Documents plugin: template-based generation can produce DOCX as well as PDF, set with outputFormat. See Generating DOCX output.
• Knowledge Base: a failed ingestion shows its status and reason, and Retry re-runs the last operation after cleaning up partial content. The Update Knowledge Base node gains a FAIL exit with per-operation rollback. See Error handling and retry.
• File Download component: end users download files directly from a rendered app. See File download.
• Global header and footer: a UI Flow can define a header and footer shown on every page. See Global header and footer.
• Event handlers: event handlers can fire on a condition. The Save Local Variable action saves the clicked collection row into local data under a key. See Conditional triggers and Save Local Variable.
• Image component: the Image component honors a color property. See Image.
• Data source testing: the database query testing modals have a Documents tab, and multi-record operations report an affected-record count. See Testing data sources.
• Mistral provider: add Mistral with your own API key under Organization Settings → AI Settings → Model Providers, then assign its models per capability. See AI providers.

​

Changed

• Documents: the Document Type field is renamed Document Tag. Behavior is unchanged.