Build a Custom Chat Experience on Your Own Data

External chat can now be driven entirely from your own application through a REST API, so you can build a branded search-and-chat experience instead of embedding the standard widget. A new endpoint lets your app discover which filters a collection supports — their fields, labels, and types — so you can render the right controls and feed the selections straight into conversational search.

Per-Key Domain and Rate Limits for External Chat

Every external chat key can now be locked down on its own. Restrict a key to specific website domains, including whole subdomain patterns, and cap how many requests a minute it will accept. A key that leaks or gets shared beyond its intended site can no longer be used from anywhere else or run up unlimited traffic. Both settings are edited per key from the key management screen.

Save and Roll Back Chat Agent Instructions

The instructions that steer an external chat agent can now be saved as named versions and restored later. Browse everything you have saved, see which one is live, and switch back to an earlier version in a single step. Tuning agent behaviour becomes reversible, so you can try a sharper set of instructions without wondering how to get back to what worked.

Restrict Agent-to-Agent Traffic to Trusted Networks

Administrators can now confine their organization’s agent-to-agent API access to an approved list of network ranges, managed from a new settings page. Add the ranges you trust, turn enforcement on or off, and remove entries as your infrastructure changes. Organizations that never configure a list are unaffected.

Create Collections and Add Documents Mid-Workflow

Workflows are no longer limited to reading collections that already exist — they can create a new collection and add documents to it as part of a run. That makes it possible to automate the whole path from gathering source material to storing it somewhere reusable, with no manual setup step first. System-managed collections stay protected from accidental writes.

Run Is Now the Primary Action in the Composer Toolbar

The Composer toolbar now leads with Run, so the action you take most often is the most prominent one. Share stays available as a compact icon beside it, and Past Runs remains reachable while a workflow is executing, making it easier to start a run and check on earlier ones without leaving the canvas.

A Clearer Main Navigation

The main menu has been reordered to follow how work actually flows, with building, running, and reviewing grouped together and data and configuration after them. Closely related areas such as Guardrails and Evaluation now sit side by side instead of being separated. The standalone page for chatting with a data collection has been retired — that conversation now happens in the Data Collections area and the Builder Agent, so there is one less place to look.

Trace WordPress Posts Back to the Workflow That Published Them

Content published to WordPress now carries the identifying details of the workflow and the specific run that produced it. Anything downstream that reads the payload can tie a published post back to its exact source run, which makes auditing what went out — and tracking down anything unexpected — far more direct.

Other Misc Bug Fixes and Improvements

Kick Off Workflows Automatically When Files Arrive

Triggers are now a top-level feature, so your workflows can start on their own the moment new content shows up — no manual runs required. Connect your own Google Drive account right in the app, choose a folder to watch, and attach the workflows that should run whenever a file lands there. You can manage everything from a dedicated Triggers page or from a Triggers section inside each workflow’s settings, making it easy to set up hands-off, event-driven automation.

A Cleaner, More Intuitive Way to Group Nodes

Grouping nodes on the canvas is now clearer and easier to work with. Group containers stay visible with a subtle dashed outline and respond as you hover and focus them, so it’s always obvious what’s grouped and what’s selected. New floating toolbars put the right actions at your fingertips — group a selection, ungroup a container, or copy, duplicate, and delete in a single click — for a smoother, more predictable canvas experience.

Other Misc Bug Fixes and Improvements

Automatic Branded Frames on Your Video Clips

A new Video Frame Overlay node composites your branded show frame — logos, banners, and wordmarks — directly onto spliced video clips as part of your workflow. It automatically matches each clip to the frame built for its aspect ratio, so nothing gets cropped or misaligned. This replaces a manual post-production step and keeps every clip on-brand without extra editing.

Publish Events, Not Just Stories, to WordPress

The WordPress publishing node can now create an event, a story, or both in a single run. A new action selector lets you choose exactly what to publish, so calendar-style content flows through the same automated pipeline you already use for articles. Existing story-only setups keep working unchanged.

Use Your Data Collections in the Builder Agent

You can now tag a data collection in the Builder Agent chat and have the agent actually use it to answer your questions. Tagged collections appear as clean, readable chips both while you type and in your sent messages — no more raw tags cluttering the conversation — and the chip clears from the input box once you send. The agent reliably pulls from the collection you referenced, so grounding a conversation in your own data just works.

See Generated Artifacts Right in Your Chat

When you ask the in-workflow chat to run a workflow, the documents and files it produces now appear as clickable cards as soon as the run finishes. Clicking a card opens the artifact in an embedded viewer alongside the chat, so you can review what was generated without leaving the conversation.

More Reliable Generation for Large Documents

Document, presentation, and spreadsheet generation nodes now handle very large inputs gracefully instead of failing when the content is too big to process at once. Oversized material is set aside for the agent to read selectively, so workflows that pipe in long documents or large amounts of text now complete successfully.

Other Misc Bug Fixes and Improvements

More Flexible WordPress Content Ingestion

WordPress data collections can now pull author content from a custom field instead of only the standard short bio, so the text that actually matters to your site gets ingested and searched. Multi-site collections also get a dedicated WordPress Settings tab where each site is an expandable card you can edit and save on its own — no more scrolling through a long stack in a cramped modal to change one site.

Other Misc Bug Fixes and Improvements

Bring Your Own Tools With My Tools

Organization admins can now self-serve add third-party tools to their workspace from Organization Settings, without any engineering help. Give a tool its connection details and optional credentials, and it becomes instantly available as a node in the Workflow Builder and to the chat agent. You can enable or disable individual tools per server and refresh a server’s available tools at any time, so your team always has exactly the capabilities you want on hand.

Group Nodes to Keep Your Canvas Organized

You can now select multiple nodes on the workflow canvas and group them into a single movable container. Drag the group to reposition everything at once while individual nodes stay editable, and ungroup at any time from the right-click menu. Grouping is purely organizational, so it never interferes with how your workflow runs and your layout is preserved across saves.

Richer Sticky Notes

Sticky notes now support inline rich text editing with a floating toolbar, making it easier to annotate and document your canvas. Choose from several font sizes and apply bold, strikethrough, bullet lists, and links directly as you type, with formatting shown live instead of behind a preview toggle. Notes grow to fit their content so everything you write stays fully visible.

Audio Output Now Appears as an Artifact

Text-to-speech nodes now produce a proper artifact card on the canvas, complete with Open and Download actions and an entry in the Files tab. Previously the generated audio could be produced but had nowhere to surface, so it was easy to miss. Now your synthesized audio shows up right alongside your other workflow outputs.

Other Misc Bug Fixes and Improvements

Faster, More Cost-Efficient Workflow Runs

We reworked how the platform handles the context sent to AI models so that repeated and multi-turn requests reuse work already done instead of reprocessing everything from scratch. This means agent-driven workflows now run noticeably faster and more efficiently, especially when you iterate on the same task or run similar steps back to back. The improvement applies across the full range of agent, research, and document-generation capabilities, so speed and cost gains show up wherever you rely on AI.

Stay on Top of Activity With the New Notification Center

A notification bell now lives in the sidebar, with an unread counter so you can see at a glance when something needs your attention. Open the Activity panel to browse your notifications grouped by time, mark items as read without leaving the page, and respond to access requests inline. A new preferences screen lets you choose exactly which notifications you want to receive.

Access the Latest AI Models

You can now work with the newest generation of OpenAI and Anthropic models, including Claude Sonnet 5. The refreshed model lineup is available both for agentic workflow runs and in the admin model picker, so you can put the most capable, up-to-date models to work right away.

Edit Published Workflows in Place

The workflow builder assistant can now modify an existing workflow directly instead of creating a duplicate. Your changes are applied to the same workflow, so existing links, shares, and run history stay intact. This makes it far easier to iterate on a workflow you have already published and shared with your team.

Other Misc Bug Fixes and Improvements

A New Home for Your Workflow Artifacts

Your workflows now have a dedicated Artifacts page where every generated output lives in one place. Browse in a grid or list, search by title, file name, or source workflow, and filter to find exactly what you need. From there you can open, preview, download, rename, share, or delete any artifact — so the things your workflows produce are easy to revisit and hand off to teammates.

Shareable Links to a Specific Workflow Run

You can now link directly to a particular run of a workflow. Open a workflow with a run already in view, and copy the URL to share or bookmark that exact run — the page restores it for whoever opens the link. This makes it much easier to point a teammate at the specific result you’re discussing.

Upload JSON Files to Unstructured Collections

Unstructured data collections now accept .json files alongside the file types already supported. You can bring structured JSON data straight into a collection without converting it first, so more of your source material is ready to use right away.

Clearer Cost and Usage Insights in the Composer Dashboard

The Composer dashboard now lets you filter token spend and output metrics by workflow label, so you can see how different teams or projects are actually using workflows, all backed by live data. The workflow list also tells the difference between a workflow that has never run and one whose latest cost simply couldn’t be retrieved — instead of showing an ambiguous dash for both — so the numbers you’re reading are easier to trust.

Workflows Run on Faster, Smarter Models

We’ve moved workflows onto Anthropic’s latest Claude Opus generation, retiring older Opus versions that are being phased out. Runs that previously used an earlier model are upgraded automatically, so your workflows keep running without interruption while benefiting from higher quality and better performance.

Other Misc Bug Fixes and Improvements

Upload Larger Audio and Video Files

You can now upload audio and video files up to 5 GB, raised from the previous 1.5 GB limit. Longer recordings, higher-resolution footage, and full-length media no longer need to be trimmed or compressed before you add them to a workflow. Other file types remain subject to their existing size limits.

Other Misc Bug Fixes and Improvements

Set Shared Instructions for an Entire Workflow

You can now add Special Instructions at the workflow level that are automatically shared with every step when the workflow runs. Set your guidance once — tone, formatting rules, domain context, or anything else that should apply throughout — instead of repeating it in each individual step. The instructions carry through on reruns and resumes too, so your context stays consistent no matter how the run started.

Export Workflow Run Metrics

Workspace admins can now export a detailed report of workflow run activity directly from the admin page. Filter by date range, status, workflow, category, owner, or label, then download a spreadsheet with per-run details plus optional cost and token usage. Summary sheets roll the results up into averages and totals, making it easy to review performance and spend across a team or time period.

Add API Keys for More AI Providers

The workspace admin panel now supports adding your own API keys for additional AI providers. This gives administrators more flexibility to bring their preferred models into the platform and manage provider access centrally.

More Reliable Document Generation

We improved the reliability of long-running document, presentation, and spreadsheet generation. Lengthy jobs are now less likely to be interrupted by transient connection drops, and when a step does hit a temporary snag it can retry and recover instead of failing outright. The result is fewer stalled or failed generations on complex work.

Other Misc Bug Fixes and Improvements

A Redesigned Workflow Details Panel

The workflow details panel has been rebuilt around a two-level view that makes it far easier to follow a run. A new Workflow Summary shows at-a-glance stat cards, alerts, and a live progress bar the moment you hit Run, and you can drill into any node to see its own stats and a full event timeline. Schedules now live directly inside the Configure tab, and a persistent header keeps the workflow and node you’re looking at in view no matter which tab you’re on.

Clearer, Filterable Run Events

The events timeline now gives you real control over what you see. A new filter menu offers quick presets, from full chain-of-thought detail down to errors and retries only, and remembers your choice. Tool calls now show their arguments and results with readable or raw JSON, one-click copy and download, and cleaner Complete, Failed, and Result labels, so tracing exactly what an agent did is much more straightforward.

Choose How Agents Handle Clarifying Questions

When an agent needs more information mid-run, you can now choose exactly how it should behave. Pick from proceeding without asking, asking and continuing after a set time, or asking and waiting indefinitely until you respond. This gives you the flexibility to run hands-off automations or keep a human in the loop for the decisions that matter, without ever losing your place in a run.

Reliable Excel Uploads With Merged Cells

Spreadsheets that use merged cells, common in real-world templates, now upload and process correctly. Previously these files could slow uploads down and hand workflows rows with missing values, leading to misaligned data. Merged cells are now flattened automatically so the values land where you’d expect and your workflows run against clean, complete data.

Finer Control Over WordPress Content Collections

Managing a WordPress content collection is now more precise and predictable. You can filter which posts are indexed by matching on specific fields, and a clear warning now appears before you remove a post type, letting you know that its already-indexed content will be dropped on the next sync. Collections also stay in step with your live site, so content removed at the source is cleaned up automatically.

More Model Choices for Document and Presentation Output

Document- and presentation-generating output nodes now offer additional fast model variants in their dropdowns, giving you more ways to balance speed and quality for your generated content. Defaults are unchanged, so existing workflows behave exactly as before while the new options are there when you want them.

Other Misc Bug Fixes and Improvements

Cleaner, More Reliable PowerPoint Outputs

Presentations generated from your workflows now render the way you’d expect. Long text no longer spills out of tight template slots like dates and footers, newly placed elements no longer collide with existing template content, and text no longer disappears into same-colored backgrounds. The result is polished, correctly structured slides that follow your formatting instructions without manual cleanup.

Faster Responses on Our Most Capable Model

Workflows and agents running on our highest-capability model now return output noticeably faster, with no change to quality. You get the same strong results in less time, making longer and more complex runs feel more responsive.

New Image Generation Model

You can now choose a new OpenAI image-generation model when creating or editing images in your agents and workflows. It sits alongside the existing options, giving you more flexibility to match the style and quality you need for a given task.

Chat Stays Put as You Explore Your Workflow

Clicking between nodes in the workflow editor no longer resets your chat. Your conversation now persists as you move around and inspect different steps, so you can explore a run without losing the context you were working in.

Search and Chat Reflect Your Latest Content

For connected content sources, posts that have been removed or unpublished at the source are now cleared from search and chat results automatically. This keeps answers grounded in what’s actually live, so outdated or deleted material stops resurfacing.

Other Misc Bug Fixes and Improvements

Add Sticky Notes to Your Workflow Canvas

You can now drop sticky notes anywhere on the workflow canvas to leave context, reminders, or explanations for yourself and your teammates. Pick a color, click to place a note, and write in Markdown with support for emojis — then minimize it out of the way or expand it in a focused editor when you need room to type. Notes are saved with your workflow and never interfere with how it runs, so you can annotate freely without affecting execution.

The AI Can Now Ask You a Clarifying Question

When your request is ambiguous, a workflow can now pause and ask you a clarifying question instead of guessing. You’ll see a clear prompt with the choices to pick from, and once you answer, the workflow continues with your intent in hand. This leads to more accurate results on complex requests and less time spent redoing runs that started off on the wrong foot.

A Cleaner, More Organized Workflow Side Panel

The workflow side panel has been redesigned to make it easier to follow what’s happening during a run. A live progress bar shows how far along your workflow is, and the events view now steps from a high-level workflow summary down into per-node detail with a clear timeline. Schedule management moved into the Configure tab and a consistent header now stays with you across every tab, so everything is easier to find at a glance.

Upload Much Larger Files in Workflows

Large file uploads in workflows are now far more reliable, with support for files well beyond the previous limits — even for workspaces that use enhanced encryption. Big documents, audio, and video that used to fail partway through now upload smoothly through a faster, dedicated path. This means you can bring substantially larger source material into your workflows without hitting a wall.

Other Misc Bug Fixes and Improvements

Edit and Splice Video Right Inside Your Workflows

A new Video AI Splicer node brings video editing directly into the workflow composer, so you no longer have to leave the platform to cut and combine clips. Connect a File Input node or paste a video URL, then describe the moments you want in plain language — for example, “scenes where animals are clearly visible” — and the node extracts the matching clips for you. The resulting clips flow through to downstream nodes and are saved alongside your other workflow files, ready to review or reuse.

See Your Workflow Outputs at a Glance on the Canvas

Generated outputs now appear as clear artifact cards directly on the Composer canvas, so you can immediately see what a workflow or node produced without digging through menus. Each card shows the file with its name and type — spreadsheets, documents, presentations, PDFs, and more — giving you an unfurl-style preview right where you’re working. This makes it far easier to understand the value of a run at a glance, even if you’re new to the product.

A Cleaner Data Analysis Node

The Data Analysis node’s configuration is now simpler, with the standalone Web Search toggle removed in favor of the more flexible tools setup. If you need the agent to reach the web, you can now configure it through tools, which keeps all external capabilities in one consistent place. Existing workflows continue to run as before.

Other Misc Bug Fixes and Improvements

Choose How Your Custom Instructions Guide External Chat

You now have precise control over how your custom instructions shape the responses in external chat and search. A new setting lets you pick between two modes: append, which layers your rules on top of the built-in defaults, and override, which lets your instructions take the lead so guidance like tone, style, and topic boundaries is followed faithfully. Your custom instructions now take clear precedence over the platform defaults, so directives you set are respected at response time. Safety guardrails always stay in place regardless of the mode you choose.

Other Misc Bug Fixes and Improvements

Transcribe Long, Large Audio Files End to End

Workflows can now ingest and transcribe much larger audio recordings, including multi-hour podcasts and radio shows in the hundreds of megabytes. Large files upload directly and reliably without stalling or failing partway through, and audio in any common format is automatically normalized before transcription so it just works. This unlocks fully automated pipelines that pull in long recordings, transcribe them, and turn the transcript into finished content.

Smarter Handling of Long Agent Runs

Long-running agent and research workflows no longer break when a conversation grows past a model’s limits. The platform now automatically condenses earlier context as a run approaches the ceiling, keeping deep research, data analysis, and document-generation nodes running to completion instead of failing near the end. The result is fewer dead-end runs and more dependable output on complex, multi-step tasks.

Choose How Custom Chat Instructions Are Applied

Admins configuring an external chat experience can now decide whether their custom instructions build on top of the platform’s default guidance or fully replace it. A simple append-or-override control makes this self-serve, with clear help text explaining the tradeoff, so teams can tune assistant behavior to their needs without engineering involvement.

Other Misc Bug Fixes and Improvements

See the Reasoning Behind Your Evaluations

Evaluations now surface the reasoning behind each result, so you can understand not just whether a check passed or failed but why. The explanation streams in live as the evaluation runs, giving you real-time insight into how each judgment is being made. This makes it far easier to trust your results, spot edge cases, and refine your criteria with confidence.

A Cleaner, More Reliable Tool Picker

We’ve tidied up the tool picker so it only shows tools that are meant to be used directly on the canvas. Previously, a number of behind-the-scenes tools appeared as selectable options, and connecting them could lead to confusing runtime errors. Now the list is focused on the tools you can actually use, for a smoother and more dependable building experience.

Other Misc Bug Fixes and Improvements

Shareable Workflow Links That Just Work

Workflow links now carry your organization context, so sharing a link with a teammate reliably opens the right workflow. If you follow a link to a workflow in a different organization you belong to, the platform switches you over automatically and lets you know — instead of showing a confusing “no longer exists” message. Older links keep working too, redirecting seamlessly to their new form.

A Live Run Timer in the Workflow Editor

The workflow editor header now features a run control that expands into a live elapsed-time timer while your workflow runs and collapses when it stops. Run status now surfaces through clear notifications rather than a crowded status bar, giving you an at-a-glance sense of progress without cluttering the header.

See Which Workflows Are Scheduled

The Workflows list now includes a Schedule column, so you can tell at a glance which workflows are set to run automatically. No more opening each workflow to check whether it’s on a schedule.

Copy and Download Agent Responses

In the workflow events panel, Agent Response blocks now offer Copy and Download controls on hover, matching the controls already available for thought-process events. This makes it quick to grab an agent’s output for reuse elsewhere.

Cleaner Reference Output Evaluation Editing

Editing a reference output in evaluations now uses a simple Text or File tab toggle instead of showing a text field and file upload at the same time. Your entered text or uploaded file is preserved until you submit, so switching between options no longer loses your work.

Upload Parquet Files to Structured Collections

You can now upload Parquet files directly into structured collections. The upload area recognizes and accepts them alongside your other spreadsheet formats, making it easier to bring in data that’s already stored in this efficient columnar format.

Faster File-Processing Runs

Workflows that process files now start up noticeably faster. Runs that handle documents, spreadsheets, and code no longer wait on lengthy setup before getting to work, so you spend less time watching and more time getting results.

Other Misc Bug Fixes and Improvements

Permanent Links for AI-Generated Images

You can now choose to have AI-generated images returned as permanent links instead of ones that expire after an hour. This means images embedded in content you publish to external sites and blogs keep displaying correctly over time, rather than turning into broken images once a temporary link lapses. Administrators can turn this on for the whole organization from the general settings page, and optionally allow it to be set on a per-workflow basis — with the existing expiring-link behavior left unchanged for everyone who doesn’t opt in.

More Resilient Generation During Provider Outages

We’ve added an automatic safety net so that generation can keep working even when the underlying AI provider has a temporary outage or is overloaded. When the primary provider is unreachable, eligible requests transparently reroute to a backup provider running the same model, so your work completes instead of failing. Everyday runs are unaffected on the happy path, and there are no changes to the quality or behavior of your results.

Review the Reasoning Behind Past Workflow Runs

The step-by-step reasoning your agents produce is now saved with each workflow run, so you can revisit it later. When you open a past run, you can read the chain-of-thought and see the agent messages and tool activity that led to the result — without having to re-run the workflow. This makes it far easier to audit, validate, and explain why a workflow produced a given output.

New “Do Not Guess” Mode for Agents

You can now turn on a grounding option on your agent steps that tells the model to stick to the evidence in front of it — your instructions, task, attached files, prior turns, and tool results — rather than filling gaps with guesses. When a fact isn’t supported, the agent will say so or reach for a tool instead of inventing an answer. It’s off by default and available across the agent, research, code, and data-analysis steps, so you can dial in extra rigor exactly where accuracy matters most.

A Clearer, Easier-to-Scan Workflows List

The Workflows list now gives titles the room they need, showing the full name on one line instead of cutting it off, so you can tell workflows apart at a glance. Descriptions wrap neatly, and the actions menu now stays pinned to the right edge as you scroll wide tables — so run, edit, and other controls are always within reach.

Other Misc Bug Fixes and Improvements

Stay on the Latest Version Without a Hard Refresh

When you keep the platform open in a tab for a long time, you may have been running an older version of the app until you manually refreshed. Now, when a new version is available, a small, non-blocking prompt appears in the corner with a Reload button so you can update on your own schedule. Nothing reloads automatically, so you never lose in-progress work — you choose the right moment to pick up the latest improvements.

Other Misc Bug Fixes and Improvements

Eval Category Descriptions at a Glance

Each evaluation category in the Evals tab now has an info icon beside its name. Hover over it to see a clear description of exactly what that category measures — the same guidance you see when setting up a new eval, now available right where you review results. No more switching context to remember what each score means.

Sharper, More Focused Evaluation Scoring

We refined how several evaluations score your content so results are more trustworthy and easier to act on. Goal Achievement now weighs goal clarity and task completion equally, focusing the score on whether the work actually met its intent. Quantitative checks were also tuned to reduce false flags, so the issues surfaced are the ones that genuinely matter.

Other Misc Bug Fixes and Improvements

More Reliable Workflow Runs Through Network Hiccups

Long-running workflows now stay connected far more reliably. The app keeps the connection alive in the background and, if it does drop, reconnects quietly without flashing an error at you. When a connection can’t be restored right away, the app checks with the server for the run’s true status before reporting anything — so a brief network blip no longer looks like a failed run, and you’ll see a clear “Offline” or “Reconnecting” indicator when your own connection goes down.

Fewer Interruptions on Complex, Multi-Step Workflows

Research-heavy and tool-intensive agent workflows can now run much longer before hitting internal limits. Previously, involved tasks that required many back-and-forth steps could stop partway through with an error. We’ve raised the ceiling substantially, so these deeper workflows now run to completion instead of aborting mid-execution.

Other Misc Bug Fixes and Improvements

Smarter Chat That Wires Up New Nodes for You

When you ask the general chat to add a node to an existing workflow, it now connects that node into your workflow automatically instead of leaving it stranded on the canvas. You no longer have to manually draw the edges after each request, so building and extending workflows through chat is faster and more intuitive.

More Reliable Evaluations on Large Documents

Evaluations that run against large file attachments like PDFs and Office documents now have more time and headroom to finish, so they complete successfully instead of cutting off partway through. This makes evaluation results more dependable when you are working with lengthy, content-heavy source material.

Other Misc Bug Fixes and Improvements

Build End-to-End Content Pipelines in the Workflow Composer

You can now assemble a complete content pipeline entirely inside the workflow composer. A new web scraping capability pulls source material straight from the web, so your workflows can gather the latest articles and pages without manual copy-paste. Combined with the composer’s generation nodes, it’s now possible to go from raw sources to finished, ready-to-review output in a single workflow.

Turn Audio Into Ready-to-Use Transcripts

A new Audio Transcription input node converts spoken audio into clean, formatted transcripts you can feed into the rest of your workflow. It automatically separates and labels different speakers, so interviews, meetings, and recordings come back organized and easy to read. You can also supply domain-specific terms up front to improve accuracy on names and specialized vocabulary.

Publish Finished Work Straight to WordPress

Workflows can now publish their output directly to a WordPress site through a new publishing node — no manual export or re-upload step. It works with both standard and custom WordPress setups, and connection credentials are entered through securely masked fields so passwords never appear in plain text. This closes the loop from gathering content to producing it to getting it live.

Schedule Workflows to Run on Their Own

You can now put your workflows on a schedule so they run automatically at the times you choose. Scheduled runs go through the same execution path as manual runs, so they behave identically — same reliability, retries, and error handling — just without anyone having to press Run. It’s ideal for recurring jobs like daily digests or regular content refreshes.

Upload Plain Text Files to Your Collections

Plain .txt files now upload cleanly to both structured and unstructured collections. Previously text files could be mistakenly treated as spreadsheets and mis-parsed; now they’re stored as the text they are. This makes it simpler to bring notes, articles, and other plain-text sources into your collections without surprises.

More Accurate, Faster, and More Consistent Evaluations

Evaluations now score your outputs more consistently and return results noticeably faster. Behind the scenes, scoring was standardized around clear, shared criteria and made more resilient to unusual inputs, so repeated runs on the same content give steadier, more trustworthy results. Multiple checks also run in parallel now, cutting the wait when a workflow has several evaluations configured.

Other Misc Bug Fixes and Improvements

Run Workflows Automatically on a Schedule

Workflows can now run on their own recurring schedule instead of waiting for someone to start them by hand. From the workflow editor you’ll find a new Schedule tab where you can set how often a workflow should run — hourly, daily, weekly, or any custom cadence — and the platform handles the rest. The workflow list also shows which workflows are scheduled at a glance, so keeping recurring jobs like periodic web scraping and content pipelines running is now hands-off and reliable.

Cleaner, Copy-Ready Evaluation Node Output

Evaluation node results now render as fully formatted output, matching the polished look of your other workflow nodes. That means past runs display cleanly instead of as raw text, and you can download the results or copy them as markdown or formatted text to reuse elsewhere. Reviewing and sharing what an evaluation produced is now consistent with the rest of the workflow experience.

Other Misc Bug Fixes and Improvements

A More Powerful Way to Evaluate Agent Output

The Eval node is now a serious tool for catching both numerical and qualitative problems in what your agents produce. New multi-dimensional evaluations score an output across distinct criteria — such as how well it achieves its goal and the overall quality of its writing — and a new Quantitative Check inspects the numbers in an output for well-formedness and consistency. When there is nothing numeric to verify, the check now clearly reports “not applicable” instead of a misleading passing score, so your results are easier to trust at a glance.

Evaluate the Files Your Agents Create

Evaluations can now read the files your workflows generate — presentations, spreadsheets, documents, and PDFs — not just inline text. This means you can hold the actual deliverable to the same standard as everything else, and get grounded, substantive feedback on the content that really ships rather than a superficial summary.

Clearer, More Reliable Eval Reviews

Reviewing evaluation results is smoother across the board. The CSV export now includes an “Upstream Node” column so you can see exactly which step each result grades, and the evaluations list stays fast and stable even when older records are involved. Together these make it easier to audit results, trace them back to their source, and keep working without interruptions.

Other Misc Bug Fixes and Improvements

Edit and Re-Run Any Past Workflow Run

Opening a past run in the Composer is no longer a read-only view. You can now edit the canvas directly from a past run — adjust node inputs, tweak parameters, rewire connections — and press Run to kick off a fresh execution from that starting point. There’s no separate “restore” step anymore, so iterating on an earlier result is now as simple as opening it, changing what you need, and running again.

Review Evaluation Results in the Sidebar

A new Evals tab in the workflow’s right sidebar puts your evaluation results right next to the canvas. Results appear as clear, collapsible cards with Pass, Fail, and Score badges that read cleanly in both light and dark mode. Selecting an evaluation node on the canvas automatically filters the tab to that node’s results, so it’s easy to focus on exactly the checks you care about.

Keep Chatting With Locked Workflows

Locking a workflow no longer shuts down the chat. Finalized workflows now keep chat available for questions, debugging, and understanding what a workflow does — while still blocking any chat-driven change that would edit the locked design. You get the best of both: a protected workflow you can’t accidentally alter, and an assistant you can still talk to.

A Clearer, Easier-to-Read Events View

The workflow Events tab has been reworked for readability. Events now show the node names you defined instead of generic internal labels, the header stays pinned as you scroll, and duplicated status text has been trimmed so each line is easier to scan. Node results group correctly and the view controls — copy, expand, and see-more — have been tidied up for a cleaner read.

Other Misc Bug Fixes and Improvements

A Rebuilt, Faster Workflow Engine

We’ve rebuilt the engine that runs your workflows to be significantly faster and more resilient. Independent steps now run at the same time instead of strictly one after another, so multi-step workflows finish noticeably sooner. Runs recover cleanly from interruptions, retry transient failures automatically, and halt immediately when a step fails so you’re never left waiting on a run that can’t succeed. Agents that rely on extended, multi-step reasoning also run more reliably from start to finish.

Pause, Resume, and Re-run Without Starting Over

You can now pause a running workflow, come back later, and resume exactly where it left off — or stop it entirely. Paused runs are held safely and pick right back up even after sitting a while. And when you re-run from a specific step, only that step and the ones after it recompute, reusing the cached results of everything upstream — so iterating on a workflow is dramatically faster and never means redoing work that’s already done.

Lock a Workflow While Keeping Inputs Open

Workflows can now be locked so two people can’t unknowingly edit or run the same one at once, protecting your work from being overwritten. Locking freezes the workflow’s logic, models, and connections as read-only, with a clear indication of who holds the lock. File and text input steps stay fully editable, so teammates can still run a locked workflow with their own files and prompts without changing how it works.

Build and Edit Entire Workflows Just by Chatting

Chat now works at the level of your whole workflow, not just a single step. Ask it in plain language to add, remove, connect, or reconfigure steps and it proposes the changes for you to review before anything is applied. The conversation stays with you as one continuous thread as you move around the workflow, making it far faster to assemble and refine complex pipelines.

A Smarter Workflow Builder Assistant

The workflow builder assistant now defaults to a more capable model, improving the quality of its suggestions and generated workflows right out of the box. You can also set custom instructions that apply to every message in your conversation, so you can tailor how the assistant works once and have it stick across sessions.

New Models: Claude Opus 4.7, GPT-5.5, and GPT-5.5 Pro

Claude Opus 4.7, GPT-5.5, and GPT-5.5 Pro are now available across agent nodes. Reach for them on your most demanding workflows to get higher-quality output on complex reasoning and generation tasks. In-product recommendations account for each model’s pricing, so you can balance capability against cost.

Automatically Upgraded to the Latest Models

Workflows still pointing at older model versions are now seamlessly upgraded to the newest generation, and you’ll see a notification whenever this happens so there are no surprises. Retired models are removed from the picker so you always choose from current options, and extended long-context support is now generally available with no special configuration.

Smarter, Cleaner File Handling

Uploaded files now flow straight into the agent and are processed securely, so you no longer have to fiddle with parser settings. Uploads are validated for type and size up front with clearer guidance on accepted formats, giving you immediate feedback instead of a failed run later. File connection points are color-coded and only accept valid files, and nodes now surface a badge when a newer version is available.

Run Your Workflows Programmatically via an API

You can now launch workflows directly from your own code or systems through a simple API — no manual clicks required. List a workflow’s saved versions, see which inputs it accepts, and kick off a run with your own values passed in on the fly. Each run accepts custom tracking identifiers, so it’s easy to embed the platform into your automated pipelines and tie results back to your own systems.

Control Who Can Access Your Data Collections

Share individual data collections with specific teammates and choose whether each can view or edit, all from a simple Share dialog. Collections stay private to you by default, and only the people you grant access to can open, chat with, or modify them. When you share a workflow, everyone on it automatically gets access to the data it relies on, so collaboration just works without accidental exposure.

Clearer, Safer Workflow Sharing

The workflow share dialog now explains what each access role can do and warns you before making a workflow broadly editable. It’s easier to understand exactly who can see or change your work, and simpler to avoid accidentally over-exposing something sensitive.

Manage Your Own Encryption Keys

Administrators get a dedicated settings area to connect and validate your organization’s encryption key vault through a guided form. A status dashboard shows your current key version, when it was last rotated, and encryption coverage across your storage at a glance. Access is restricted to designated administrators, giving your security team direct control over data-at-rest protection.

Add Whole Tool Groups to Your Canvas in One Drag

Related tools are now grouped together, so you can drag an entire set onto your canvas in a single motion instead of wiring each one by hand. Multi-select lets you grab several at once, and newly dropped nodes connect automatically — turning a repetitive setup chore into just a few clicks.

Cleaner, Right-Angle Connections on the Canvas

Workflow diagrams now support an alternative connection style that routes links along clean right angles instead of curves, and you can flip between the two layouts anytime from the canvas zoom controls. Connections match the color of what they link and route neatly around nodes instead of cutting through them, so your workflows are easier to read at a glance.

More Predictable Node Selection

Selecting and expanding nodes in the tool picker now behave independently and predictably: clicking a row selects it, the chevron expands it, and clicking again deselects. Building workflows feels more precise and less fiddly.

A More Powerful Admin Dashboard

The admin dashboard now shows total cost in dollars alongside run counts for your most-executed workflows, and the Model Distribution view is back so you can see how your agents are spread across different models. Refreshed charts and clearer labels make your usage and spend easy to understand at a glance.

Friendlier Names for Your Tools

Tools now show readable, descriptive names and helpful descriptions instead of raw technical identifiers, so it’s much easier to understand what each one does while building a workflow. This applies across both built-in and connected third-party tools.

Deploy Shareable Apps Instantly

Apps you deploy are now publicly accessible by default, so anything you publish is immediately shareable without extra setup. When you need to restrict access, a single toggle turns authentication back on — making quick demos and internal sharing far faster while keeping control in your hands.

Clean Up Your Chat History in One Click

You can now remove individual conversations directly from the chat sidebar — hover any conversation, delete it, and confirm to permanently clear it and its messages. If you delete the conversation you’re currently in, the view resets cleanly to a fresh chat so you never lose your place.

Other Misc Bug Fixes and Improvements

“My Workflows” Now Includes Shared Workflows

The “Show only my workflows” filter now surfaces both the workflows you created and the workflows that have been shared with you. Previously the filter only matched workflows you personally created, so anything a teammate invited you to was hidden. Now a single toggle gives you everything you’re actually working on.

Other Misc Bug Fixes and Improvements

Find Your Workflows Faster With Clearer Filters

The workflow dashboard now gives you two independent filters instead of a single toggle: My workflows shows only the ones you created, and Shared with me shows the ones teammates have given you access to. Turn on either or both to focus your list, and leave them off to see everything. Your filter choices are saved in the page address, so you can bookmark or share a filtered view and return to exactly what you were looking at.

Other Misc Bug Fixes and Improvements

Deep Research That Uses Your Connected Data

A new agentic deep-research node can now run multi-step research directly against the data and tools you’ve connected to the platform, instead of relying on the open web alone. It plans, gathers, reflects, and synthesizes across several passes, calling the tools available to it as it works. This lets you point rigorous, iterative research at your own sources and get grounded, tool-backed answers.

Build Prompts by Chatting With Your Node

You can now refine a node’s instructions through a conversation right inside the workflow editor. Instead of writing and rewriting a system prompt by hand, you can chat back and forth to shape exactly what you want the node to do. It turns prompt-writing into a guided dialogue, making it faster to get agents behaving the way you intend.

Conversational Workflow Builder

A new chat-based workflow builder lets eligible organizations assemble workflows by describing what they want in plain language. The assistant helps put the pieces together for you, lowering the barrier to creating useful automations. It’s rolling out behind a controlled flag so teams can adopt it at their own pace.

Enter Task Text Alongside Connected Inputs

You can now type instructions into a node’s task field even when an input node is already connected to it. Previously the connected input could override or disable your typed text; now both are read together and combined into the final request the agent receives. This makes it much easier to add specific guidance on top of the content you’re feeding in.

Organize Workflows With Labels

Workflows can now be grouped and categorized with labels that your organization defines. Admins can create, color, edit, and remove labels, anyone can tag their own workflows, and you can filter the workflow list by label to quickly find what you need. It brings order to a growing collection of workflows and makes shared spaces far easier to navigate.

Sharing Controls: Default Access and Revocation

Sharing got more flexible and more governable. Workflow owners can now set who can view, run, edit, or share their work, apply a sensible default level of access for their organization, and — new in this release — revoke someone’s access at any time. Revoked users get a clear, friendly message rather than a confusing error, so collaboration stays controlled without being disruptive.

Don’t Run Workflows With Disabled Steps

If a saved workflow contains a node or tool that has since been turned off for your organization, the platform now stops the run up front and tells you exactly which steps are the problem. Previously these stale workflows could fail unpredictably or produce unexpected results. You now get clear, actionable feedback instead of a confusing failure.

Self-Serve Usage and Consumption Dashboard

Administrators now have a self-serve dashboard for visibility into usage and token consumption across their organization. It surfaces consumption over a date range you choose, recent trends, a breakdown by model, and the top consumers — all in one place. This makes it easy to understand and manage how your organization is using the platform without waiting on a report.

Pan the Canvas With the Middle Mouse Button

In the workflow editor’s selection mode, you can now press and hold the middle mouse button to pan around the canvas without switching to hand mode. It’s a small but welcome shortcut that keeps you in flow when you’re moving between selecting nodes and navigating a large workflow.

Stronger Protection for Data at Rest

Stored files and data can now be transparently encrypted at rest using application-level envelope encryption, adding an extra layer of protection beyond standard storage safeguards. This happens automatically behind the scenes with no change to how you work. It’s part of our ongoing investment in keeping your data secure.

Other Misc Bug Fixes and Improvements

Import and Export Workflows

You can now export a workflow and bring it into another environment, making it easy to reuse and share your work across teams and setups. When you import a workflow, it’s automatically checked so anything that needs your attention is surfaced up front — you’ll be told if uploaded files need to be re-added, if a step isn’t available in the current environment, or if a field is missing a value. That means fewer surprises and a clear path to getting an imported workflow running.

Other Misc Bug Fixes and Improvements

Evaluations Refreshed With the Latest Judge Models

Evaluations now run on an up-to-date lineup of judge models, including the newest GPT and Claude releases, while older, deprecated options have been retired. Automatic evaluation setup is working again too: name and describe what you want to measure and the platform will draft your pass criteria, fail criteria, and judge settings for you. Choosing a current judge model now creates your evaluation smoothly instead of failing with an unexpected error.

Other Misc Bug Fixes and Improvements

Choose Which Tools and Nodes Your Team Can Use

You can now decide exactly which tools and workflow nodes are available inside your organization. From a new Nodes & Tools settings area, admins can turn individual capabilities on or off, use per-category switches, or enable and disable everything at once. Changes are staged so you can review them before saving, making it simple to tailor the platform to how your team actually works without waiting on us to make the adjustment for you.

Other Misc Bug Fixes and Improvements

Single Sign-On With Your Identity Provider

You can now sign in to the platform using your organization’s own single sign-on, starting with Microsoft Entra. Just enter your work email and you’ll be routed straight to your company’s login — no separate password to manage. This makes access simpler for you and easier for your administrators to govern centrally.

Share Workflows and Control Who Can Do What

Workflows now have built-in ownership and sharing. You can invite teammates to a workflow and grant exactly the level of access each person needs — from viewing and running to editing and sharing with others. This makes it easy to collaborate on shared workflows while keeping sensitive work protected.

Clearer Messages When the AI Service Is Busy

When the underlying AI service is temporarily overloaded or unavailable, you’ll now see a clear, friendly message telling you it’s a temporary upstream issue and to try again in a moment — instead of a confusing raw error. This applies across document, spreadsheet, presentation, and data-analysis generation, so it’s easier to tell when a hiccup is momentary rather than a problem with your work.

Other Misc Bug Fixes and Improvements

Renamed Deep Research Agent

Renamed Deep Research Agent to Web Deep Research to better reflect its functionality.

Claude Sonnet 4.6 Support

Added support for Claude Sonnet 4.6, now available for selection across the platform.

Auto-Generated Names for Required Nodes

Nodes with mandatory name fields (e.g., Evals) now automatically receive a default name to prevent confusion and ensure a smoother user experience.

Improved Events Bar Performance

Fixed an issue where long content in the events bar rendered and scrolled slowly. Content now loads and scrolls smoothly without delayed animations.

Tool Naming

Updated the Procured Data and Unstructured Data tool names to be more concise and easier to read.

Powerpoint Updates

Removed “Upload Template” functionality from the PPT node.

Implemented improvements to address PPT workflow behavior and reliability.

Data Tool Updates

Azure Search (unstructured data) remains available, with some nodes removed for simplification.

Removed data indices selection from unstructured data nodes.

UI Cleanup

Removed the Advanced Settings section from the UI, as these settings were not functional. This reduces confusion and ensures only supported features are visible. Existing nodes remain unchanged to avoid disrupting current workflows.

Node Input Method Update

We’ve simplified how nodes receive input to make workflows more consistent, predictable, and easier to understand.

As part of this update, node settings are now managed in a configuration panel that opens when you select a node. This panel contains standard settings like toggles, menus, and non-connected fields.

What Changed

Previously, some node parameters could be set in two different ways:

• By typing directly into a field
• By connecting another node into that parameter’s port

Only one method could be active at a time, which often caused confusion. Now, parameters that accept dynamic input are handled exclusively through ports.

If a node has a port for a value, that value is expected to come from another node (for example, a Text Input node or an Agent node). Node settings live in the configuration panel, while inputs now come only through node connections.

In short:

Ports are for inputs. The configuration panel is for settings.

How This Affects Existing Workflows

Existing workflows will continue to work.

For workflows created under the previous system, we’ve added legacy support for text fields that were used instead of a connection.

If one of these fields contained text in your workflow:

• The field will still appear in the configuration panel
• It will be labeled as deprecated
• A note will explain that this input can now be provided using a Text Input node

If the parameter was already connected to another node, nothing changes.

Updating Old Workflows (Optional)

Workflows using legacy fields will continue to function. However, you can update them to the newer model at any time.

Recommended approach:

• Add a Text Input node
• Enter your text there
• Connect it to the parameter port on the node
• Delete all text from the associated field in the configuration panel (this will make the field disappear)

Editing a legacy field:

• If you edit the text in a legacy field, the workflow will use the updated text
• If you delete all text from the field, the field will disappear. At that point, input must be provided using a connected node

Nodes with Legacy Input Fields

The following nodes previously allowed direct text entry for inputs that are now port-based. If text was entered in older workflows, that field will still appear in the configuration panel with a deprecation note.

• Guardrails — Query
• Evaluations — Query
• Structured Data — Query text
• Unstructured Data — Query text
• Agent — Task
• Data Analysis Agent — Task
• Deep Research — Deep research query
• PowerPoint — Task
• D3 Data Visualization — Task
• Text to Speech — Transcript
• Spreadsheet — Task
• Word Document — Task
• Network Graph — Task
• Text Redaction — Input data

What’s New & Improved

• Updated platform design deployed with a refreshed UX for a cleaner, more consistent experience.
• Procured data is now available directly within tools, reducing setup friction.
• Chat support added for PowerPoint, Excel, and Word nodes, making it easier to iterate on document outputs in context.
• Prompt Enhancer now works for text inputs, helping users quickly refine prompts before running workflows.
• Uploaded PowerPoint templates are applied correctly, including styles and layout patterns.
• Fixed a looping issue in PowerPoint workflows where nodes could get stuck waiting for user input instead of progressing.
• Node outputs are easier to discover — the output button is now positioned above the node, reducing the need to search via hover or extra clicks.
• Agent names now stay in sync between the node modal and the sidebar.
• Users can view outputs from all past runs, not just the most recent execution.
• Past runs are locked by default when reopened, preventing accidental edits unless explicitly unlocked.
• Chat guidance improved to prevent missed prompts — required agent input now appears directly in chat, reducing timeouts.
• “Run complete” messaging is more accurate and no longer appears mid-execution.
• Structured CSV uploads now work reliably through the Data Collections tab.
• Tools within agents are now categorized, making them easier to browse and understand.

Platform Cleanup & Enhancements

• Removed SQLite data tools that are not relevant for current use cases.
• Removed or hidden non-functional or irrelevant platform elements to reduce clutter.
• Manual MCP button removed from the platform.
• “No Client Data” messaging made more prominent to reinforce data handling expectations.
• Basic Admin screen added for foundational administrative visibility.

New Capabilities & Model Updates

• Claude Opus 4.6 deployed across the platform.
• Data Analysis agent and document output nodes now use Claude Opus 4.6 by default.
• Automatic context compaction enabled for Data Analysis and document output nodes when using Opus 4.6.
• File input parameters are now supported across all document generation nodes, allowing files to be attached similarly to the Data Analysis agent.
• Azure Search (Unstructured Data) tool deployed, enabling richer enterprise data retrieval.

Composer

• Categorized Tools: Workflow tools now organized into intuitive categories for easier discovery
• Past Runs Dropdown: Access and review previous workflow executions with full history
• Unsaved Changes Warning: Modal warning before leaving when you have unsaved changes
• Chat Integration: Interactive chat capabilities integrated into workflow editor
• Chat Tweaks: Multiple improvements to chat functionality and UX

Collection Chat

• Collection Chat Improvements: Multiple enhancements to chat interface and functionality
• Reset Text Input: Reset functionality when resetting collection chat
• WordPress Rules: Enhanced WordPress integration rules and configurations

Bug Fixes

Various bug fixes across workflows, evaluations, guardrails, data sources, UI components, and dependency updates to improve stability and user experience.

Collection Chat

• Collection Chat Interface: Full chat interface for unstructured Qdrant collections
• Markdown Formatting: Assistant messages render with proper markdown formatting
• Chat Component: New reusable chat component for workflow integration

Workflow Enhancements

• Drag from Input to Create: Drag from any input node to create and connect new nodes
• MCP Server Modal: New interface for managing Model Context Protocol server configurations
• Improved Auto-Zoom: Better auto-zoom behavior when adding new nodes to canvas
• Query Length Validation: Added validation for unstructured collection search queries

Data Sources & Collections

• Chat Integration: Chat interface for unstructured collections
• Multi-Select Component: Enhanced multi-select with search functionality
• Script & Database Files: Accept uploading script and database file types

Node Types

• PPTX Export: Export PowerPoint presentations from slides node

Bug Fixes

Various bug fixes across workflows, data sources, guardrails, evaluations, UI components, and performance to improve stability and user experience.

Workflow Productivity Features

• Undo/Redo System: Full undo/redo support for all workflow operations (Cmd/Ctrl+Z)
• Cross-Tab Copy/Paste: Copy nodes from one workflow and paste into another workflow tab
• Multi-Node Selection: Select and manipulate multiple nodes simultaneously
• Play/Stop Control Bar: Enhanced execution control bar with real-time status

Workflow Advanced Features

• Drag Selection: Dynamic node positioning with drag selection and center focus
• Connection Line Improvements: Enhanced visual feedback when connecting nodes
• Workflow Metrics: Track execution time

Search & Discovery

• Cross-Page Search: Unified search functionality across multiple pages
• Filter Search Component: Advanced filtering for better search results

Data Sources

• Hybrid Search: Combine keyword and semantic search for better retrieval accuracy
• Select All Collections: Bulk operations with select-all functionality
• Source Weight Slider: Adjust search result weighting with visual slider controls

Support Features

• Raise Support Tickets: Submit support tickets directly from the platform
• Prompt Enhancer: AI-powered prompt enhancement to improve prompt quality

UI Components

• Doc Viewer Component: Display and render documents in workflows
• Markdown Table Rendering: Improved markdown table display
• Tako Node Display: Carousel-style iframe display for special nodes

Bug Fixes

Various bug fixes across workflows, data sources, evaluations, guardrails, UI components, and security to improve stability and user experience.

WordPress Integration

• WordPress Collection Configs: Full WordPress API integration for content ingestion
• WordPress Support: Connect to WordPress sites as data sources via API

Node Creation & Workflow Enhancements

• Create Node from Output: Click output ports to spawn new connected nodes
• Human Intervention Node: New node type that pauses workflow for human input/approval
• Unstructured Data Node: Query unstructured collections directly in workflows
• Multiselect Node Parameters: Nodes can now have multi-select input parameters

Node Execution & Monitoring

• Stream Node Results: Node results stream in real-time during execution
• Node Status Highlight Halo: Visual halo effects highlighting active node execution status
• Preview Scrolling: Fixed and improved preview panel scrolling behavior

LLM Playground

• Saved Prompts: Save and reuse frequently used prompts
• Default Model Persistence: Last-used model remembered across sessions

Data Source Improvements

• Dataset Quality Metrics: View quality scores and metrics for your datasets
• Primary Color Theming: Support for organization-specific primary color branding
• Preview Badge: Visual badge indicating preview/sandbox environments
• MIME Type Handling: Enhanced file type detection and labeling
• Override Guardrails: Ability to enable/override guardrails per data source

Chat Improvements

• Chat Interface Enhancements: Multiple improvements to chat functionality
• Responsive Iframe: Full-height responsive chat display
• Text Streaming: Fixed text streaming and message grouping
• Chat Message Await: Support for awaiting user input in chat workflows

Bug Fixes

Various bug fixes across workflows, data sources, search functionality, and UI components to improve stability and user experience.

Composer

• Dark Theme Support: Full dark mode support for workflow editor with proper styling
• Workflow Publishing: Publish workflows for sharing and reuse across organization
• Node Explorer & Search: New sidebar for browsing nodes with search functionality
• Inspector Sidebar Improvements: Enhanced inspector with better node configuration UI
• One-Click Workflow Access: Simplified workflow opening from dashboard with single click

Data Sources

• Soft Deletion System: Collections marked for deletion are retained for recovery period
• Revert Deletion: Restore accidentally deleted collections before permanent removal
• Data Sources UI Refresh: Redesigned data sources interface with improved navigation
• Collection Page Improvements: Enhanced dropdown menu with refresh functionality and better loading states
• Advanced Search with Filters: Multi-criteria search across collections

Evaluation System

• Split Configuration & Dashboard: Separate pages for evaluation config and results dashboard
• Polling for Results: Real-time polling for evaluation completion status

WebSocket & Real-Time Updates

• Stream Logs from WebSocket: Real-time log streaming during workflow execution
• WebSocket Stability: Multiple fixes for connection handling and reconnection logic
• Result Streaming: Get workflow results directly from WebSocket connection

Slide & Chat Panels

• Chat and Slide Deck Panel: New panels for interactive chat and slide presentations
• Chat Message Support: Foundation for chat message handling in workflows

Node & Template Improvements

• Node Status in JSON: Node execution statuses now tracked and saved in workflow JSON
• Template System: Save and load workflow templates for reuse
• Slide Node Support: Support for PowerPoint slide generation node
• File Upload Node: Fixed and improved file upload capabilities in workflows

Bug Fixes

Various bug fixes across workflows, data sources, UI components, and evaluation features to improve stability and user experience.