New features and improvements shipped every week.
The newest Claude models are here, and agent nodes now pick the most capable model by default.
Claude Sonnet 5 is now available in the builder and in the Claude Code node model picker
Agent nodes now default to Claude Opus, so new agents start with the strongest model
Multi-tenant API integrations get a dynamic tenant picker, letting one credential serve all your tenants
Workspace owners can delete their workspaces again
A big week for integrations. BuildOps, CompanyCam, WorkMax, XOi, MyGeotab, and Foundation Software all join the catalog as Decisional-managed toolkits.
BuildOps: 293 endpoints covering the full field-service management surface
CompanyCam: 62 endpoints for photo documentation and project tracking
WorkMax: 27 read tools for time tracking and workforce data
XOi: 22 tools for field technician workflows
MyGeotab: 38 read tools for fleet telematics, with cursor pagination on every entity
Foundation Software: construction accounting, ready to connect with a credential
Come back to any app and keep building. App Builder now opens a fresh session against an existing app, and approval gates can show approvers exactly what they need.
Reopen any existing app and continue building in a new session, no matter which workspace you're in
Approval gates can now present the exact fields an approver needs to see, both in the workflow panel and on approval pages
Integration toolkits connect automatically the first time the builder uses them, instead of stopping you at a gate
Apps are now a first-class product. Build interactive dashboards and tools that live on their own, with no agent needed to serve them.
Describe the app you want and Decisional builds it: hosted, shareable, and independent of any single agent
Published apps render from a dedicated pipeline with hardened sandboxing
Long runs no longer stall when they hit a provider rate limit, and failures now tell you what actually happened.
Runs now ride through provider rate limits instead of stalling
Run failures are now classified by their real cause (rate limit, auth, provider error) instead of a generic failure state
Self-healing alerts now explain why a repair was attempted
Fixed partially failed runs not routing to auto-repair when retries were involved
Build status no longer regresses when late build events arrive after completion
Create an account with just your phone number. Get a code over SMS, sign in, and add an email later if you want.
Phone-number accounts on SMS, with code-based login, auto-registration, and an optional email completion step
Uploaded files now come through reliably as tool attachments in agent runs
See your whole workspace's AI usage on the dashboard, clone agents across workspaces, and explore interactive dashboards your builder generates.
The dashboard now rolls up AI usage and tool stats across every member of your workspace
Clone agents across workspaces, so a working agent can move to another team without a rebuild
Live artifacts (beta): builders can generate interactive dashboards you can open and explore, not just static outputs
Build your own integration toolkit right from the operator surface. Describe the API, and Decisional generates, tests, and registers the tools for you.
Custom Toolkit Builder: point it at an API and get a working toolkit, scoped to your org
The settings page got an overhaul, with cleaner sub-routes, collapsible sections, and paginated tables
Agent cards and the list view now show who created each agent, and you can filter the agents list by creator
Claude Fable 5 joins the builder and agent-node model pickers
Talk to your agents from Microsoft Teams, and manage your workspace with simpler roles backed by a full audit log.
Microsoft Teams support: connect your agents to Teams channels the same way you do Slack
Workspace roles are now just Owner and Member, with governance controls for who can see and edit what
A new audit log records every user action. Setting changes show before and after values, with deep links to live agents
Webhook triggers now accept file uploads end to end, and tool-action approvals go through even if you approve hours later.
Send a file to a webhook trigger and it flows through the whole run
Approvals are now executed by a durable worker, so a late approval still lands reliably
A new "Connect with Claude" button links your own Claude account without pasting a key
Channel actor credentials for SMS and email, so channel messages act as the right user
Restart a failed run from where it stopped instead of from scratch, and choose which Claude model builds your agent. You can even switch mid-thread.
Resume failed and partially completed runs from the failed node, with all upstream work preserved
Pick your Claude model in the builder and switch mid-thread, with inline model selection in the route picker
Bring your own Claude account, with attribution, health checks, and failure policies
Agent email and Slack threads can now wake a running workflow node for human-in-the-loop replies
Workspaces now come in four flavors: personal, private, public, and protected. You control exactly who can find and join each one.
Four workspace visibility types: personal, private, public, and protected
Auto-created personal workspaces are named after their owner
New native integrations for Motion (project management) and Billtrust Invoice Gateway
Tool-action approvals let you require a human sign-off on any tool call, not just gates
Agent sandboxes can now stretch to two hours when they need it, and new users get a guided onboarding flow from their first prompt.
Sandbox runs extend on demand from 25 minutes up to 2 hours, so long automations finish instead of timing out
A new onboarding flow guides first-time users from signup to a working agent
The agent manager gains a state store tab where you can inspect the durable state your agent keeps between runs
Sketch your workflow on an Excalidraw canvas and let Dex build from it, plus new runtimes including Cursor Composer 2.5 and Claude Opus 4.8.
"Draw a diagram" in the prompt box menu opens an Excalidraw canvas. Sketch the flow you want and Dex builds from it
Cursor Composer 2.5 joins the code-agent runtime lineup
The builder and Claude Code nodes are upgraded to Claude Opus 4.8
A new Linq messaging gateway brings your agents to SMS, photos and media included, alongside a new ServiceTitan integration.
Message your agents by text through the Linq SMS gateway, with media attachments in both directions
ServiceTitan native integration joins the catalog
Agent cards now show when each agent was last active
Composio credentials refresh automatically instead of expiring silently
Log in to Decisional with your Slack account, and tune node settings on live agents without taking them offline.
Sign in with Slack: one click from your workspace, no separate password
Node settings can now be updated on live agent versions without a redeploy
Sidebar hierarchy and thread controls are polished for faster navigation
Self-healing now repairs each failed node individually when a run partially fails
Agents can now publish reports and dashboards as hosted HTML artifacts, real URLs you can open and share.
Agents publish reports and dashboards to a real URL you can share
The workspace orchestrator gets its own email inbox with realtime updates and thread deletion
Live browser previews stream right into the run view, so you can see the page your agent is on while it works. Office files preview without a download too.
A live browser preview renders inside node execution details, streaming the page your agent is on mid-run
Word, Excel, and PowerPoint outputs open in the sidebar preview
Agents can re-authenticate expired browser sessions on their own
Every agent can now have its own email inbox. It receives requests, replies with real attachments, and keeps each conversation in its own thread.
Give an agent an email address. Incoming mail wakes it, and replies go out with real attachments
The inbox view groups messages by email thread, with sender policies to control who can write in
A new thinking level setting lets you dial how deeply your orchestrator reasons
See your workflow take shape without leaving the conversation, and everything that used to be in Pro is now included in Free.
The workflow preview renders inside chat mode as Dex builds, and you can jump into any node from it
Pro plan features are now included in Free
Pending questions moved into the composer, so answering Dex feels like replying instead of filling out a form
Pasted snippets render as markdown with inline previews
Node-scoped integrations show up directly in the workflow panel
A guided demo walks new users through their first build, and Dex chat gains file attachments with a live streaming indicator.
A hands-on onboarding demo that builds a real agent with you
Attach files in Dex chat, with previews and a streaming indicator while Dex responds
Dex now streams its thoughts and tool calls into chat in real time as it builds. Brex also joins as a Decisional-managed integration.
Dex streams its thinking and each build step into home chat live, so you can follow along instead of staring at a spinner
Brex joins as a Decisional-managed toolkit, so you can connect spend data with no API key setup
Replace a single credential key without redoing the whole connection. Existing keys are shown, and the form adapts to each credential type
Slack conversations keep one transcript per channel instead of forking a new session on every thread reply
Define custom integrations scoped to your org, with your own toolkits and your own tools. Plus Telegram bots for talking to your agents.
Org-scoped custom integrations: define your own toolkits and tools, stored and versioned in Decisional
Talk to your agents from Telegram with a per-org bot connection
Fan-out concurrency caps let you throttle how many branches run at once on any edge
A cleaner workspace with consistent backgrounds and tabbed settings, plus credits now meter and resume per node so a long workflow never stalls silently when you run low.
Workspace surfaces refreshed with consistent backgrounds, pill-style tabs, and a tabbed settings panel
Pick your runtime right from the home build prompt and the single-panel build view
Integrations catalog now focuses on connecting one toolkit at a time, with clearer setup steps
Node-level credit metering: workflows pause and resume cleanly when credits run low instead of failing mid-run
When a run fails, Decisional now diagnoses the cause and either fixes it itself or asks you for the missing detail on Slack. Your run resumes with the answer instead of starting over.
Self-healing for failed runs: Decisional triages the failure, patches and retries code-level bugs automatically, retries transient errors, and escalates anything that needs a human decision
Slack clarification: when the only blocker is a missing rule, mapping, or threshold, Decisional DMs the workflow owner with the run context, applies your answer, and retries the failed node
Reminder pings at 60 minutes, 2 hours, and 24 hours so a stalled run doesn't get forgotten; the repair session expires after 24 hours
Runtime picker redesigned as a logo dropdown, available across build modes
A Decisional-managed filter layer for Composio triggers narrows incoming events down to just what you care about
Switch any agent between Chat, Operator, and Build modes from a single picker, without losing your thread.
Mode switcher per agent: flip between Chat, Operator, and Build from one dropdown. Threads pick up where you left off when you switch back
Unified promptbox across modes so the input stays consistent no matter what the agent is doing
Smart thread resumption: switching modes mid-conversation keeps your context intact instead of starting fresh
A pinned chat overlay on public run pages, an in-app AI assistant for authenticated viewers, and a tighter workflow viewer that keeps run context where you need it.
Public run chat overlay pinned to the right of the page so stakeholders can ask questions about a run without leaving it
AI Assistant chat overlay for authenticated viewers, so you can get explanations and next steps without bouncing to a separate thread
Workflow viewer keeps run-context inside the viewer and fuses node tabs into a single Configurations panel
Pinned section in the sidebar with toolkit tiles, plus a visited-agent bump so your most recent work stays close to hand
Cleaner credit pricing, a single Dashboard for Usage and Tool Calls with per-user attribution, and tighter approval handling.
Credit pricing rebased: simpler math, same value
Growth plan now includes 75,000 credits per month
Usage and Tool Calls merged into Dashboard tabs with per-user attribution and account type columns, so you can see exactly who spent what
Approvals hardened: prevents double-approve, adds a bypass toggle for solo runs, and surfaces pending state more clearly
A denser, faster sidebar with a unified Sessions list, plus a Chat mode in the home promptbox so you can pick up where you left off without hunting through pages.
Unified Sessions list in the sidebar: every recent thread, run, and build in one collapsible section with faded edges so the rest of the nav stays clean
Denser nav with collapsible session groups so long histories stop pushing things below the fold
Chat mode toggle in the home promptbox: flip from "Build an automation" to a direct chat without leaving the page
Use Decisional-managed toolkits without bringing your own API keys. Pick the platform credential, and pay for usage via credits.
Exa search is now a platform toolkit: no account setup, no key paste. Usage is metered via your credit balance
New Platform credential option in the credential dropdown for any toolkit that supports it. Drop it in like a regular credential
Pause any workflow for human review with a new Gate node, then share a public approval link so stakeholders can sign off without needing a Decisional account.
Add a Gate node anywhere in your workflow to pause execution until a human approves or rejects the next step
Set quorum rules on each gate — require sign-off from any single approver, or wait for every approver on your list
Share a public approval link so stakeholders can see the full context and approve or reject without logging in to Decisional
Query your data warehouse directly with the new Snowflake integration, and move through large Bill.com lists with full pagination support.
Connect Snowflake as a data source to query your warehouse from any agent
Paginate through large Bill.com lists to pull in records beyond the first page
Fixed Snowflake credential setup so pasted private keys (PEM) are accepted correctly
Fixed Bill.com PDF document downloads failing for some invoices
A cleaner shimmer sweep replaces the old glow on active workflow nodes, with matching dark-mode polish across the canvas.
A subtle shimmer sweep now marks running workflow nodes instead of the old pulsing glow, so it's easier to spot what's in-flight at a glance
Dark-mode polish across the workflow canvas so node states, edges, and selection highlights are consistent end-to-end
See exactly what each node is doing as it runs. Live execution logs stream directly onto workflow nodes so you never have to guess what's happening inside a run.
Watch live execution logs stream onto each workflow node as your agent runs — see tool calls, results, and errors in real time without leaving the canvas
Structured log rendering groups tool calls and results into clean, scannable entries instead of raw text
Copy the full log output for any node with one click for easy debugging and sharing
Keep moving while your assistant is still replying. Line up the next prompt, adjust the order, and send it the moment you're ready.
Queue your next message while the assistant is still streaming so you can keep working without waiting for the current reply to finish
Reorder or remove queued follow-ups before they send so your most important request goes next
Interrupt the current response and immediately send the next queued message when priorities change
Search the manual integration picker to find the right connection faster
See timestamps on chat messages and on build, run, and test cards for clearer context
Move through a cleaner sidebar with better hierarchy and sharper editor polish across the workspace
Open a searchable command menu with Cmd/Ctrl+K to move through your workspace without digging through tabs and sidebars.
Open a searchable command palette with Cmd/Ctrl+K to jump straight to the action you need
Jump directly to Instructions, Workflow, Worksheet, All Runs, and Outputs for the current agent from one place
Open Files, Skills, Integrations, and Triggers without hunting through sidebars and nested panels
Use single-key shortcuts like B for Agent Manager and L for Assistant after opening the palette
Navigate globally to create a new agent, open All Agents, check the Dashboard, or manage workspace integrations
Approve workflow steps from Slack, and get guardrails before going live.
Approve or reject workflow steps directly from Slack DMs — when a node needs your sign-off, you get a message with full context and one-tap Approve / Reject buttons
Trigger validation catches missing required fields before a live run starts, so runs no longer fail silently on bad input
Warnings surface when a trigger could fire on every event without filtering, so you can add guards before going live
Clearer thought labels show exactly which tool your agent is using during a run instead of generic "Contemplating"
Browse outputs in folders, open spreadsheets inline, and label your credentials.
Browse agent outputs in a redesigned folder view with nested navigation, file previews, and one-click downloads
Spreadsheet outputs now open directly as workspace tabs instead of popups — click any spreadsheet artifact to view it inline
Label your credentials (e.g. "Work" vs "Personal") so your agent knows exactly which account to use
A live status indicator shows when your workflow is rebuilding so you always know what's current
Fixed tool connection badges showing stale status after switching agent versions
Fixed the workflow viewer sometimes staying pinned to an older run instead of the latest one
Fixed shared credentials not being recognized, causing unnecessary re-connection prompts
A simpler setup flow — build your agent first and connect integrations when you're ready.
Build your agent first, connect integrations later — no more blocking setup steps before you can start building
Fixed the arcade game locking you out of other tabs during a build
Filter your dashboard by time range, and enjoy a more reliable experience across the board.
Dashboard now supports date-range filtering — view metrics for 24 hours, 7 days, 14 days, 30 days, or a custom range
Filter sidebar on the agent listing page for faster navigation
"Run Now" properly inherits trigger configuration so manual and scheduled runs behave the same
Fixed stale node details and approvals panel requiring a page refresh
Fixed approval cards reappearing after being dismissed
Fixed attachments being silently dropped in the assistant prompt box
Improved agent templates, Excel file support, and a more accurate workflow viewer.
Improved agent templates with better sharing, browsing, and cloning experience
Upload and process Excel files directly in your agent workflows
Workflow viewer now displays conditional branching (decider) nodes
Workflow viewer shows the actual graph from historical runs so you always see exactly what ran
Clicking a workflow node during a run opens its details alongside the run panel
Play a game while your agent builds, and enjoy faster startups across the board.
Play a mini arcade game while your agent is building
Watch your workflow run step-by-step in a live viewer during test runs
Test runs are clearly labeled so you never confuse them with live data
Build progress shows real-time logs so you can see exactly what's happening
Agents start up nearly instantly thanks to pre-warmed environments
Resuming an agent is 15–20 seconds faster
Upgraded to a more powerful AI model for better accuracy
Better organization for skills, and a smoother connection experience.
Skills and knowledge base files now have separate upload dialogs so it's clearer what you're adding
Skills your agent creates during a run are automatically saved for future use
New built-in email-writing skill
Connecting an integration now happens right inside the chat thread as an interactive card
Clicking "Connect" takes you straight to the sign-in flow — no extra steps
You get a warning before deleting an integration that's being used by an agent
A context usage bar shows how much of your agent's memory is being used, updated in real time
Your agents can create PDFs, and you choose which steps need your sign-off.
Agents can generate and export PDF files during runs
Toggle "requires approval" on individual nodes in your workflow
A dedicated approvals panel shows all nodes waiting for your sign-off
Scroll through previous conversation history with your agent
Browse, preview, and download every file your agent produces.
Browse agent output files organized in virtual folders
Preview JSON and text files directly in the sidebar without downloading
Download any file your agent produced with one click
Approve actions faster with bulk actions, AI-generated previews, and a stacked card design.
Approve all pending actions in one click instead of going through them one by one
Each approval card now shows an AI-generated preview of what the agent is about to do, so you know exactly what you're approving
Pending approvals stack with a 3D depth effect so you can see how many are queued up
A smarter prompt box that helps you get started with animated suggestions.
The prompt box now shows animated placeholder suggestions with integration icons to help you get started
More visibility into runs, team invites, and faster agent wake-ups.
Run details now show a step-by-step breakdown of everything that happened
Download the actual files your agent produced during a run
Invite members from your organization directly into your workspace
Granola integration is live
When a tool connection fails, you see the actual error instead of a generic failure message
Agents hibernate instead of shutting down, so they wake up much faster when you come back
Two new node types let your workflows tap into AI reasoning and your knowledge base directly.
New AI Node that runs AI-powered processing, summarization, extraction, or classification inside any workflow node
New RAG Node that retrieves relevant context from your uploaded documents and knowledge bases
RAG automatically chunks large files, generates embeddings, and stores them in a vector index so queries return only the most relevant passages, even across hundreds of pages
Supports PDFs, Word docs, CSVs, and plain text with overlapping chunks to preserve context across section boundaries
Give your agents specialized knowledge with Skills, build agents entirely through conversation, and enjoy a smoother experience.
Upload markdown skill files to give agents specialized knowledge, instructions, or playbooks — with preview and raw viewing modes
Upload files directly to your agent during chat conversations for processing
Building and testing agents now happens through natural chat messages instead of separate buttons — the entire agent creation feels like a conversation
Redesigned home page with a clean centered layout and collapsible sidebar
Action nodes in test runs now pause before executing, giving you full control to approve or reject each action
Agents automatically recover from inactive sandbox states — conversations always work, even after extended breaks
Browser notifications alert you when agent builds complete
Higher character limits in the prompt input box for longer, more detailed agent instructions
Clearer thought labels show what your agent is doing (e.g., "Searching Web", "Fetching Page")
New animated sandbox setup loader while your agent's environment initializes
Switch between agent versions, copy Agent IDs for debugging, and enjoy real-time schedule updates.
New version selector lets you switch between agent versions and restore previous configurations
Edit live agents without breaking them - changes are saved as a new draft until you're ready to publish
Copy Agent IDs directly from the agent page header for easier debugging and API integration
Schedule changes sync instantly across all your open tabs - no more refreshing to see updates
Your agents now catch configuration errors before they happen, with automatic validation of integration tools.
Automatic tool validation prevents errors by checking that all referenced integrations exist before building
Clearer error messages show you exactly which tools are available when something goes wrong
Trigger configurations now support custom input schemas for more flexible automation
Request access to workspaces with a single click, and use slash commands for faster navigation.
Request access to workspaces you don't have permission to view - workspace owners get notified instantly
New slash commands let you quickly navigate and take actions without leaving your keyboard
See who's requesting access to your workspace with names instead of cryptic IDs
Chat with your agents from Slack DMs, and get deeper visibility into how your agents run.
Sync Slack DM threads with your workspace - chat with agents directly from Slack
See a thinking indicator while your agent processes messages from Slack
Rich Block Kit messages make agent responses more interactive and easier to read
Filter runs by Test vs Live to quickly find what you're looking for
Live badge shows which runs are currently executing in real-time
View detailed input and output data for each node in your workflow
Developer mode shows the Python code behind each workflow node
Copy workflow code and resize the code panel to fit your screen
Download debug JSON for troubleshooting complex workflows
Trigger your agents automatically when events happen in your connected apps.
Set up integration triggers to run agents when events happen in Slack, HubSpot, and more
Configure trigger settings with a new dynamic form that adapts to each integration
Credentials are now locked when your agent goes live, preventing accidental disconnections
Integration logos appear in color throughout the app for easier recognition
Major improvements to the agent building experience with enhanced auto-scroll, redesigned action cards, and a smoother setup wizard.
Enhanced auto-scroll behavior in the AI assistant panel for a more natural conversation flow
Redesigned action cards with clearer visual hierarchy and better touch targets
Improved setup wizard flow with progress indicators and contextual help
New ways to visualize and manage your agents with Kanban boards and intelligent workflow layouts.
New Kanban board view for the agents listing page with drag-and-drop support
Intelligent node positioning for hierarchical workflow visualizations
Integration toolkit logos now display on workflow nodes
Agent runs pagination with status filtering
Scheduled triggers visible directly in the workflow viewer
Visual feedback when AI edits your spreadsheets and improved workspace management.
Beautiful shimmer animation shows when AI is actively editing spreadsheets
User worksheets now open as tabs instead of side panels
Enhanced workflow viewer with better node positioning and integration display
Redesigned integration connection flow with spotlight overlay
Improved integration management and new starter templates.
Click-to-scroll, hover highlighting, and auto-expand for toolkits in connection panel
Action message history for Build, Test, and Connect actions
10 new agent templates for common automation patterns
Flexible data management with auto-created worksheets and full user worksheet support.
Auto-creation of worksheets with full user worksheet support
Better file upload experience with progress animations
Tabbed interface and automated agent scheduling.
New tabbed interface with VS Code-style performance optimizations
Agent scheduling system for automated runs on your preferred cadence
Run Now button for immediate agent execution
Real-time run updates via WebSocket
File upload directly in the initial prompt box
Node title, description, and type visible in job execution panel
Watch your agents being built in real-time.
Real-time build logs stream to your screen as agents are built
New content sections with beautiful WebGL-enhanced visuals.
Full dark mode across the entire app — toggle between light and dark themes from any page
Templates section with pre-built agent templates
Setup wizard moved to bottom of chat thread for cleaner flow
Start building intelligent agents that work for you. No code required.
