August 31, 2026
**Friction-free storage activation:** extended compatibility with legacy S3-compatible providers, provisioning available straight from Diagnostics, and configuration verified at startup.
- Improvement: Broader provider compatibility
- Improvement: Provisioning from Diagnostics
- Improvement: Configuration verified at startup
August 31, 2026
**Document storage hardened:** an independent security review of the new document subsystem, with every relevant issue closed before production activation.
- Security: Concurrency under control
- Security: Personal upload sessions
- Security: Remote operations faithful to their provider
- Improvement: Guide kept aligned
August 31, 2026
**Documents join your processes:** contracts, reports and attachments are uploaded straight from forms, with versions, preview, quotas and a grace period — on S3-compatible cloud storage governed by FlowSharp.
- Feature: Document field in your processes
- Feature: Documents section in Settings
- Security: Grace period and controlled deletion
- Improvement: Per-tenant storage quota
- Integration: S3-compatible storage with no vendor lock-in
- Feature: Guide: Documents and Cloud Storage
August 28, 2026
**Your installed base gets itself in order:** the new Diagnostics finds work stuck on never-configured automations and unblocks it, privacy protections survive copies and versions, and item creation speaks the language of your process instead of JSON.
- Feature: Published flow diagnostics
- Security: Privacy protections survive copies and versions
- Improvement: A history that stands up to audit
- Feature: New Item with a guided form
- Improvement: Daily work without friction
- Improvement: Calendar and events built for volume
- Integration: More predictable APIs
- Design: An interface with room to breathe
August 28, 2026
**Items get a real title and AI plays by your rules:** the title is visible everywhere, searchable and composable from a template — even AI-generated once data is complete; empty automations no longer claim success; and what reaches AI models is decided per flow, protected by default.
- Feature: The item title, everywhere it matters
- Feature: Title template, optionally AI-generated
- Security: AI receives only what the flow allows
- Improvement: Empty automations no longer claim success
- Improvement: Deleting a Process Map unwires everything, and says so
- Improvement: Distinguishable task runs and a searchable filter
- Improvement: AI summaries honest for integrations too
- Improvement: Procedures with preview and rich media
- Design: One bar across Monitor and Analytics
August 27, 2026
**Reorganized settings, guides one click away, and summaries that admit their age.** The Management section becomes Settings with a readable order, every page gains a question mark that opens the right guide chapter in its own window, and the AI Summary stops presenting itself as current when something important has happened in the meantime.
- Design: Settings, reorganized
- Feature: The guide where you need it
- Improvement: An old summary no longer looks fresh
- Improvement: Changing a task type no longer blocks saving
August 26, 2026
**Process times tell the truth and stuck phases become visible.** "Time per status" no longer mixes completed stays with ongoing ones, and phases where cases are stuck — instead of disappearing from the panel — now show how many they are holding. Plus: actions that name the process they act on, a localized canvas, and the place where agents will live.
- Improvement: Reliable time per phase at last
- Improvement: Deleting a map really unwires its processes
- Design: Actions name what they act on
- Design: A localized canvas
- Feature: Where agents will live
August 26, 2026
**Sensitive data stays sensitive and the numbers tell the truth.** An indicator that would reveal one person's data now stays hidden for everyone, impossible durations no longer distort time and cost — and you can see how many were excluded — and processes that wait for several tasks to finish now publish without workarounds.
- Security: Indicators on sensitive data protected from every role
- Improvement: Time and cost without impossible durations
- Feature: Processes waiting on several tasks now publish
- Security: Tighter control over outbound connections
- Design: A hidden value says it is hidden
August 25, 2026
**Rules now see the whole flow:** "wait until every task is done" patterns finally work, placeholders resolve in every task type, event-driven processes publish without false alarms — and the guide, accessibility and error messages take a leap forward.
- Feature: Rules across the whole flow
- Improvement: Placeholders resolve everywhere
- Improvement: Event-driven processes without false alarms
- Improvement: The guide lands on the right paragraph
- Improvement: Accessibility: names, keyboard and plurals
- Improvement: Error messages without duplicates
- Design: Polish: logo and map panels
August 24, 2026
**The flow list scales and diagnostics tell the truth.** /flows loads families in pages with server-side search; Data diagnostics flag unresolvable references, default rules carrying conditions and Sub-flows without a destination (with no more false alarms); a rejected save shows a clear banner with phase and rule; procedures and guides get real Markdown typography.
- Improvement: The flow list, paged
- Improvement: Data diagnostics tell the truth
- Improvement: Save errors you can see
- Improvement: Procedures you can read
- Security: Stricter import and forms
August 24, 2026
**A wave of tester-driven fixes.** Flow import applies the same rules as the designer, time metrics become credible again, the flowchart PDF really shows the arrows without wasting pages, clicking an activity opens its card, and an accessibility pass makes Profile, OAuth revocations and the MCP tool catalog readable with a screen reader too.
- Improvement: Import with the designer's rules
- Improvement: Credible time metrics
- Improvement: The flowchart PDF, complete
- Improvement: The click opens the right activity
- Improvement: Accessibility: Profile, revocations, MCP catalog
August 23, 2026
**Documentation becomes the source of truth again.** A brand-new guide chapter on the Flowchart, every chapter realigned with the real product, technical references regenerated from the code and first-administrator setup fixed: what you read now matches what the product does.
- Feature: The Flowchart gets its own guide chapter
- Improvement: The whole guide realigned with the product
- Improvement: Setup: the first administrator can actually be created
- Improvement: Technical references regenerated from the code
August 23, 2026
**From the flowchart to the Procedure.** One click and the AI drafts a chaptered procedure — one per phase, with the activities, who performs them and the expected result, automatic phases included — which you review and save in ActiveSOP, already linked to phases and activities. Phases now carry their own guide, visible to whoever works on the item.
- Feature: Convert to Procedure
- Feature: Linked from the first save
- Feature: The phase guide
- Security: References always within the tenant boundary
August 23, 2026
**The Flowchart becomes a document.** Download the flowchart as PDF — cover, paginated diagram with cross-page connectors, legend and the "Who does what" cards of every phase — and the people who run the process open it read-only instead of the old map.
- Feature: Flowchart as PDF
- Feature: The Flowchart for the people who run the process
- Improvement: Who does what, always explicit
- Improvement: The flow description in the canvas
August 23, 2026
**Quick fix for the Flowchart.** The "Expected output" field of tasks now saves as expected: in 0.85.0 saving the flow rejected it as an unknown field.
- Improvement: Expected output that saves
August 23, 2026
**The Flowchart.** Next to the designer comes the documentation view of the process: phases laid out automatically, large boxes with their activities inside, who does what and the expected result of each activity, in three layouts of your choice. The first step of the "process documentation" track.
- Feature: Flowchart with three layouts
- Feature: Who does what, at a glance
- Improvement: Same inspector, same structure
- Feature: Expected output with AI suggestion
- Design: It's called Sub-flow
August 23, 2026
**The security wave.** Three doors closed in a single pass: webhook responses can no longer drag sensitive data into internal records, AI client registration gets real boundaries — size, quota and lifecycle — and a hostile PDF can no longer slow the platform down.
- Security: Watertight webhooks
- Security: Boundaries for AI client registration
- Security: Hostile PDFs take no hostages
August 23, 2026
**The interface polish wave.** Twelve refinements in a single pass: dark mode becomes readable everywhere, unread notifications truly stand out, the compact sidebar breathes, and a series of details — favicon, autofill, reveal eyes — stops misbehaving.
- Design: Dark mode readable everywhere
- Improvement: Unread notifications you can actually see
- Design: Tidier compact sidebar
- Design: Layout without rough edges
- Improvement: Details that no longer get in the way
August 22, 2026
**Errors that tell you what to fix.** The Designer fork rejects reusing a request with different content, import errors list duplicate keys and the phases involved, export refuses documents with broken references, and OAuth errors speak the OAuth contract.
- Improvement: Actionable import errors
- Security: Fork idempotency without surprises
- Improvement: Fail-closed export and OAuth
August 22, 2026
**The cycle's final hardening.** OAuth errors with the right code, administration of dynamically registered clients with deactivation, traced client metadata diagnostics, automations that reject never-interpolated templates, and an MCP section that calls things by their name — with a state readable by screen readers too.
- Integration: OAuth client administration
- Improvement: Automations without illusions
- Improvement: The MCP section tells it like it is
August 22, 2026
**Metrics that declare what they measure and an audit trail that says who acted.** Every dashboard number explains its basis, unit and population; alongside structural automation comes the period's observed automation; the trend shows since when data exists. The audit trail records who acted and through which channel, and alert telemetry reflects real deliveries — with one less personal data item in payloads.
- Improvement: Explainable numbers
- Improvement: Audit with clear identities
- Security: Honest alert telemetry and less personal data
August 22, 2026
**Process maps that don't lose pieces and health that tells the truth.** Deleting a process wired into a published map is now blocked with the list of affected maps; a node whose process is gone gets flagged; archived maps are truly deleted when nothing uses them; a never-checked map no longer looks healthy; and the map list gains real search and pagination.
- Improvement: Published maps are protected
- Improvement: Honest map health
- Improvement: True cleanup of archived maps
- Improvement: Map list with real search and pages
August 22, 2026
**Lists that don't lie and AI assistants with two more tools.** Changing page really skips a whole page, totals stay stable while you scroll, search reaches the task calendar, and counts respect your visibility. MCP assistants gain task listing and data updates with the same rules as the APIs, and read-only users are no longer offered impossible actions.
- Improvement: Reliable pagination and totals
- Integration: AI assistants list tasks and update data
- Security: Counts within your perimeter
- Improvement: Read-only without mirages
August 22, 2026
**Truly immutable versions and data protection that follows the family.** Reactivating a version used by items creates a new draft instead of reopening it; an archived process can't be republished or shared by mistake; duplicate copies exactly the chosen version; and a new sensitive classification on the parent process also protects child elements already generated in other processes.
- Improvement: Reactivations without surprises
- Improvement: Archived means archived
- Improvement: Honest settings on published versions
- Security: Sensitive data protection reaches children
August 21, 2026
**What you save is what runs.** Designer rules, events and forms are verified at the source: a default rule can no longer hide a condition that would be ignored, an "on event received" rule must point to an actually subscribed event, an import no longer creates nameless event definitions, and form schemas are checked against what the runtime can interpret. Plus, Human phases gain the Rules tab for event-driven waiting.
- Improvement: Event-driven waiting for Human phases too
- Improvement: Unambiguous rules: default and conditions don't mix
- Improvement: Event rules point to real events
- Improvement: More honest imports for events and forms
- Security: No more internal errors from empty event references
August 21, 2026
**Events delivered all the way through and clear answers even in edge cases.** The event engine always executes every action of an event, in the right order and without duplicate effects; an overly wide calendar range is rejected with an explicit message, deleting an initiative from the Plan never leaves orphaned elements, and integrations see more while revealing less.
- Improvement: Every event action reaches its target
- Improvement: Calendars without silent narrowing
- Improvement: Always-consistent deletions from the Plan
- Integration: Agents also see process events
- Security: More discreet authorization errors
August 21, 2026
**No more pieces lost along the way.** An imported or saved process arrives whole or is rejected with the exact list of what to fix, and retries on element creation can no longer return the wrong result.
- Improvement: All-or-nothing imports and saves
- Improvement: Smarter import of portable documents
- Security: Safe retries on element creation
August 21, 2026
The Designer warns you before, not after. Non-editable versions open read-only, events are always chosen explicitly, and real-time diagnostics flag impossible links, unreachable outcomes, and unsupported forms before they become a problem.
- Improvement: The canvas respects the lifecycle
- Improvement: Preventive diagnostics on events
- Improvement: Forms validated at save time
August 21, 2026
Sensitive data protected across all versions, transparent AI Summary. Sensitive-field classification applies to a process's entire version family, history shows what was actually protected, the AI Summary communicates clear states, and integrations get new tools with explicit risk labels.
- Security: Sensitive fields protected across the whole version family
- Improvement: Transparent history on applied protection
- Improvement: AI Summary with clear states
- Integration: New MCP tools with risk labels
August 20, 2026
MCP profiles with tailored permissions, duplicate-free events. Each integration can have its own profile with a dedicated address and explicit permissions, OAuth authorization respects the profile's boundaries, and event delivery is protected from duplicates even when things go wrong.
- Feature: MCP profiles with dedicated addresses
- Security: OAuth authorization within the profile's boundary
- Security: Explicit permissions for MCP keys
- Improvement: Duplicate-free events, even after a hiccup
- Security: Public links revoked with the process
August 20, 2026
Remote connectors now truly connect with OAuth. The authorization flow introduced in the previous release only partially worked: remote clients could not find the discovery endpoints. Now they can, and the What's New page is easier to read on phones.
- Integration: OAuth discovery reachable for remote connectors
- Design: What's New page easier to read on mobile
August 19, 2026
OAuth-ready MCP connectors, more transparent events. Remote connectors can authorize without keys in URLs, versioning preserves event actions, and telemetry now distinguishes receiving an event from actually applying an effect.
- Integration: OAuth for remote MCP connectors
- Improvement: Versions that preserve event actions
- Improvement: Event outcomes based on real effects
- Security: Consistent Process Map scope and idempotent requests
- Improvement: More predictable imports and lists
August 17, 2026
More reliable events, clearer rules. The "deadline first, then the event" rule now covers every delivery scenario, event-linked actions are thoroughly checked before going live, and SLA durations respect a uniform maximum limit.
- Improvement: Even more reliable event delivery
- Improvement: Event actions thoroughly checked
- Improvement: Clear limits on SLA durations
August 17, 2026
A security-focused update. Two-step verification for administrators, secure one-time alert confirmations, stronger and rotatable at-rest encryption, sign-in links and keys kept out of logged URLs, and less sensitive data retained in history.
- Security: Two-step verification for administrators
- Security: Secure one-time alert confirmation
- Security: Stronger, rotatable encryption of sensitive data
- Security: Sign-in links and keys kept out of logged URLs
- Security: Less sensitive data kept in history
- Improvement: Stricter filters and searches
- Integration: Safer integrations by default
August 15, 2026
Clearer rules, alerts that actually arrive. One draft per version family, events and timeouts applied in the order they really happened, SLA warnings delivered to owners through alert policies, and an export that flags plain-text credentials.
- Improvement: One draft per version family
- Improvement: Events and deadlines in real order
- Feature: SLA warnings delivered to owners
- Security: Export flags plain-text credentials
- Improvement: More predictable search and filters
August 14, 2026
The 0.83 release is now more robust in real-world cases. Retries, versions, timeouts, and SLAs produce reliable outcomes; export and duplication preserve the design; Events gains recovery tools; compare and Analytics costs reflect the same runtime reality.
- Security: Retries and completions without duplicates
- Improvement: More faithful export and duplication
- Feature: Observable and recoverable Events
- Improvement: Complete comparisons between versions
- Improvement: Reconciled Analytics costs
August 13, 2026
Flows without hidden dead ends, operational recovery under control. The designer blocks processes that cannot reach a final state, automation stalls become visible alerts, cross-install invitations survive retries and crashes, Admins can recover an item through MCP, and the extension downloads preconfigured from Settings.
- Improvement: The designer sees real exit routes
- Improvement: Automation stalls become actionable
- Security: Federation invites resilient to replay and forwarded links
- Integration: Administrative MCP break-glass
- Feature: Extension ready from Settings
August 12, 2026
Edit a published flow without losing anything, and an engine that always tells the truth. "Save" on a published flow now creates a new version with all your changes, in one step. Sensitive data follows the data wherever it is copied, engine failures become visible and recoverable, the Events page is complete, and the operational views show what is really happening.
- Feature: New version without losing work
- Security: Sensitive data follows the data
- Improvement: Visible failures, reliable outcomes
- Feature: Events: guided activation and observable delivery
- Improvement: Operational views that tell the truth
- Improvement: Maps follow versions
August 9, 2026
Your flows travel, your work stays safe. Duplicate a flow with one click, import it from file, paste or URL, share it across installations with an expiring link. And every change made from tabs, guides or services is now visible to the designer save: no more silent overwrites.
- Feature: A flow that truly travels
- Feature: Duplicate and import three ways
- Feature: Share across installations
- Security: URL import with network guards
- Improvement: Everyone's work, visible to everyone
August 9, 2026
A smaller, verifiable distribution. No feature changes: the backend now contains production dependencies only, migrations have a dedicated image, and every release publishes image inventory and provenance.
- Improvement: Clean builds, atomic releases
August 8, 2026
Every channel, the same rules. UI, API, import, AI generation, agents and child flows now write into the engine through the same funnel: one grammar for durations, identical validations everywhere, event schemas enforceable on demand, explicit data inheritance in child flows. And the browser extension gets a makeover, in your brand colors.
- Feature: One funnel for every write
- Improvement: Durations and timeouts without ambiguity
- Feature: Enforceable event schemas, explicit inheritance
- Design: The browser extension gets a makeover
August 8, 2026
Reliable versions, honest baselines. Creating flow and Process Map versions is now concurrency-proof — no more duplicate numbers, even under bursts — and past duplicates are cleaned up for good. Savings proposals declare when the design changed underneath them, with guided regeneration.
- Improvement: Concurrency-proof versioning
- Improvement: Baselines know when they are stale
August 7, 2026
Time tells the truth. A scheduled item no longer looks "already started": the start date appears only when the process actually begins, and metrics, SLAs and sorting follow suit. This release also brings real per-visit analytics, a rule validator with no code execution in the browser, and more predictable API contracts.
- Improvement: The start date is a fact, not a forecast
- Improvement: Real per-visit analytics
- Improvement: Explicit flow lifecycle and predictable lists
- Security: Validation without execution, diagnostics without data
August 7, 2026
Real governance on published flows. A published flow is now immutable — you change it by creating a new Draft version — and each item's history records the exact design version behind every event. Misconfigured rules are stopped at publish time, AI agents follow the same policies as the UI, and Process Maps gain full link management.
- Feature: Immutable published flows, faithful history
- Improvement: Configurations stopped before they become problems
- Improvement: Agents and UI, same rules
- Design: Process Maps: links under control
August 7, 2026
Four targeted closures. Activity sequences now resume reliably from every channel — including AI agents via MCP —, completed items no longer accept changes even under concurrency, the scheduler no longer produces false conflicts, and the activities' technical configuration stays on the server. No migrations: a simple upgrade.
- Improvement: Sequences guaranteed from every channel
- Improvement: Completed items truly closed, scheduler without false conflicts
- Security: Activity configuration stays on the server
August 7, 2026
Honest responses, sequences that never stall. Second stabilization wave: concurrent saves have a single winner, activity sequences always resume — even after a crash —, scheduled items start when you actually move them, event rules now work on human phases too, and the n8n contract says the same thing everywhere: UI, snippets and guides. No migrations: a simple upgrade.
- Improvement: Sequences always resume, from every channel
- Improvement: Concurrent saves with a single winner
- Improvement: Scheduled items: they start when you move them
- Feature: Event rules now work on human phases
- Security: Historical outcomes redacted in rules too
- Integration: A consistent n8n contract everywhere
August 6, 2026
The crash-safety guarantees become complete. An independent audit combed the previous release for remaining gaps — and this release closes them all: deliveries that can no longer be lost, durable resumes, no superseded executor able to overwrite fresh work, and dangerous configurations blocked before they reach production. No migrations: a simple upgrade.
- Improvement: Guaranteed deliveries and crash-proof resumes
- Improvement: The designer blocks configurations that would get stuck
- Security: Protected identities and per-tenant boundaries
- Improvement: Saving Challenge: accrued means closed months only
- Improvement: Reopenings always tracked
August 5, 2026
The engine becomes crash-proof. Seventeen coordinated changes close critical races and crash windows: events that can no longer be lost, outcomes with a single winner, always-recoverable creations, atomic escalations. With one declared breaking change for n8n integrations.
- Improvement: No lost events, no duplicate outcomes
- Security: Hardened event provenance and alerts that really arrive
- Integration: n8n callbacks with an explicit contract
- Improvement: Always-honest operational health
August 4, 2026
Foundations refreshed. A focused update of the database access layer, shipped on its own: cleaner transaction behaviour under load and ecosystem security patches. No change to everyday use.
- Improvement: More resilience under load
August 4, 2026
Security hardening. Five fixes from an independent security review, all on the boundary between different workspaces and on what ends up in technical logs. No change to everyday use: these are barriers tightening.
- Security: Boundaries between workspaces
- Security: Logs and permissions
August 4, 2026
Nine decisions closed, no grey areas. The ambiguous behaviours accumulated over the past weeks — who receives an event, what a script knows about the child flow it created, when an item is genuinely concluded, how views are ordered at equal priority — were decided one by one and implemented together.
- Improvement: Events reach the right recipient
- Improvement: Scripts, child flows and honest conclusions
August 3, 2026
The work-item lifecycle, with no blind spots. Events raised by activities and scripts now genuinely fire and start the listening flows, an unreachable child flow is stopped at publish instead of remaining a promise, every outcome — succeeded, failed, skipped — appears in the item's history, and a failed activity can be retried, skipped or resolved with a tracked note. Plus three security hardenings from an independent external audit.
- Feature: Events genuinely fire
- Feature: Outcomes in the history, and a way out of failures
- Improvement: Child flows are no longer promises
- Improvement: Configurations stopped before they mislead
- Security: Hardenings from the external audit
August 2, 2026
What the flow declares now genuinely holds. Waiting for data, work-in-progress limits, response times, required fields on completion, priority: all of these could be configured, and then nobody — wholly or partly — enforced them. Waiting now genuinely suspends the phase's work and clears the moment the data arrives, a queued item no longer moves as if it were active, escalation creates the work in the destination phase, and required fields apply identically on every channel.
- Improvement: Waiting for data is a real wait
- Improvement: The work-in-progress limit is a limit
- Improvement: Response times produce effects
- Improvement: Required fields: the same rules on every channel
- Improvement: Priority orders the views
- Improvement: Entity reference and uniqueness are two separate choices
- Improvement: Settings that no longer mislead
August 1, 2026
Automatic rules now do what the designer promises. A rule written exactly the way the help, the hints and the assisted generator show it saw no data at all and stayed silent forever, with no error and no trace. Rules now genuinely read the item's data, numeric comparisons understand numbers that arrive as text, a field never filled in no longer sends anything forward, and when a rule does not fire the history says why.
- Improvement: Rules finally read the item's data
- Improvement: Nobody moves forward without being entitled to
- Improvement: The history says why a rule did not fire
- Improvement: Saves that lose nothing
- Improvement: A phase deadline no longer restarts
- Improvement: An empty filter no longer errors the list
August 1, 2026
Confidential data stays confidential everywhere, and everyone sees only what is theirs. A safeguard is worth as much as its least guarded exit: a field could be hidden on the item's page and appear in the clear in the indicator computed from that very field, and a read-only role could see no flow in its own list yet still learn names and volumes from another screen. Confidentiality now follows the data wherever it is recomputed, visibility is the same from whichever direction you arrive, and connections with external assistants work the way the standard expects.
- Security: Confidential data stays confidential everywhere
- Security: Everyone sees only what is theirs
- Integration: External assistants and credentials
- Improvement: Searches and long fields no longer bring the server down