> For the complete documentation index, see [llms.txt](https://docs.lleverage.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.lleverage.ai/changelog/readme.md).

# Product updates

A full rundown of the latest releases, improvements and fixes in Lleverage

{% updates format="full" %}
{% update date="2026-11-16" tags="new-releases,improvements,fixes" %}

## Week 47: a new navigation, and a closer look at your data

{% embed url="<https://www.youtube.com/watch?v=r5TlsVe8sIY>" %}

Every project now has one navigation: your pinned pages at the top, then one list of the chats and requests that need you. Builders reach every page of the project from one place. And the numbers on your pages open up: click one to see the records behind it.

### New features

* **Chats and requests in one list.** The foot of the navigation holds one list with three tabs: **Needs me** for the requests waiting on you, **Chats** for your conversations with the Agent, and **Completed**. An orange dot marks something new, and a request opens full page and reads like a chat, for example a missing PO number with the form to fill it in. Pinned pages keep their place above it, and **New** starts a chat. [Navigation](https://docs.lleverage.ai/run-operations/overview/navigation)
* **Every builder page in one place.** Builders get **Configure** at the top of the navigation. Point at it to see every page of the project grouped as Automations, Context, Monitoring, Integrations and Settings; **View all** opens the Configure page, with one line on what each page is for and **Search this project** across workflows, tables and pages. [Navigation](https://docs.lleverage.ai/run-operations/overview/navigation)
* **Click a number to see its records.** On a live widget, click a big number, a bar or a row and a side panel lists the records behind it, such as the six orders from Kuipers Hout waiting on a PO. The panel lists every condition that made the number (**Widget filter**, **Page** and **Clicked**), so the count matches, and a record opens with one click. [Live widgets](https://docs.lleverage.ai/run-operations/overview/live-widgets)
* **Steer the Agent while it works.** A message you type while the Agent works waits its turn below the conversation, with **Send now**, edit, copy and discard. **Send now** steers it into the work under way, for example a new delivery date, and the Agent carries on with it. **Stop** stays in view while you type, and the chat marks where you stopped. [Chat and sessions](https://docs.lleverage.ai/run-operations/agent/chat-and-sessions)
* **Approve with every detail in view.** When the Agent asks to change something in a connected system, such as creating a sales order in Exact Online, the approval lists every value it will send: the customer, your reference, the dates and the lines, with readable labels and dates. Secrets are never shown. [Use the Agent](https://docs.lleverage.ai/run-operations/agent/getting-started)
* **Copy a workflow to another project.** **Copy to project…** in a workflow's menu on the Workflows page puts a copy in another project of your organisation as an unpublished draft. It then lists what the copy still points at that the other project doesn't have, such as a data table or a connection, so you can point those steps at the right ones before you publish. [Build a workflow](https://docs.lleverage.ai/build-and-improve/automate/build-a-workflow)

### Improvements

* **Call the Agent again.** When a voice call ends, the chat shows where, with its length and **Call again**. During a call, a soft glow behind the message box shows who is talking, and mute shows a struck-through microphone. [Call the Agent](https://docs.lleverage.ai/run-operations/agent/call-with-agent)
* **View larger.** A table that's too wide or longer than ten rows gets **View larger**, which opens it full screen, in the Agent chat, on pages and in requests.
* **Archive or delete a chat.** Chat menus have **Archive** next to **Delete**, a right-click on a chat opens the same menu, and letters work as shortcuts while a menu is open.
* **A model for each schedule.** An Agent schedule runs on the model and effort you pick, and the Agent picks a lighter model for routine checks. The model picker is simpler: choose Claude or GPT, then slide the effort. A chat a schedule started shows a small clock. [Schedules](https://docs.lleverage.ai/administer-and-govern/control/schedules)
* **Approve from Teams.** Approvals, choices and connection buttons from the Agent arrive in Teams as a card with buttons, as they do in Slack. A document the Agent makes in a group chat or channel is also offered in your personal chat, to save to OneDrive. [The Agent in Microsoft Teams](https://docs.lleverage.ai/run-operations/agent/agent-in-microsoft-teams)
* **Alert groups you can manage.** In Organisation Settings, Notifications, **Alert groups** have **Edit** and **Delete** and list the alerts that use them; the bell on an activity view manages them too. [Alerts](https://docs.lleverage.ai/build-and-improve/monitoring/alerts-system)
* **Pages by role.** A role can be limited to specific pages, as with tables and workflows, and everyone opens a project on the top page they can see. Admins can view the project as another role after switching on **Show role switcher** under My Account.
* **Where a widget's data comes from.** The line under a widget names its data table, such as **From Orders**, and opens the workflows that fill it and when they last ran. [Project overview](https://docs.lleverage.ai/run-operations/overview/project-dashboard)
* **Clearer widgets.** A widget with nothing to show says "Nothing yet for this period", charts name every bar and fill every day, and number cards share one layout.
* **Read-only workflows.** Someone whose role can view but not edit workflows opens a workflow read-only, with its versions, runs and tests.
* **Runs that explain themselves.** A failed run that can't be retried says why, a trigger request that was refused shows in Activity as **Not started**, and a connection that lacks a permission names it in plain words.
* **More.** The connection link page and email say what the connection will do and list the permissions the provider will ask for, a bounced link email shows as such, a new connection suggests a first thing to ask the Agent, and new users choose between a tour and getting straight to work.

### Fixes

* Outlook attachments download when the setting was never saved, and file names with accents download with their real name.
* Long text in a live widget cell is cut instead of left out, and a request table keeps its rows when a cell is outside its choices.
* Moving a pinned item in the navigation no longer reloads the page, and Gmail search returns whole To and Cc addresses.
  {% endupdate %}

{% update date="2026-11-10" tags="new-releases,improvements,fixes" %}

## Week 46: pages the Agent builds with you

{% embed url="<https://www.youtube.com/watch?v=W83CK6uB2RM>" %}

This week is about the pages you run the work from. Describe a page and the Agent builds it, with widgets of its own design when a standard chart won't do, and you can change any page by hand. Working with the Agent gets more direct too: point at a row, paste a screenshot, and get drafts you can copy or send.

### New features

* **Describe a page, the Agent builds it.** Pages has its own place in the navigation, between Intelligence and Monitoring: the Overview first, then the project's own pages. **New page** asks what the page is for, for example "orders waiting on a PO, handling time per day and orders per customer this month", and the Agent names it, connects the data and builds it beside you, checking every widget before it hands it over. [Project overview](https://docs.lleverage.ai/run-operations/overview/project-dashboard)
* **The Agent builds custom widgets.** When a standard chart doesn't show what you need, such as a heat map of late orders per customer and weekday, the Agent draws the widget itself in HTML, in the Lleverage look. A widget can read the other live widgets on its page, and the Agent changes it with small edits when you ask. [Live widgets](https://docs.lleverage.ai/run-operations/overview/live-widgets)
* **Change a page by hand.** The pencil at the top of a page switches edit mode on. Move, resize and remove widgets, add one with the widget builder, and change the page's title and description in place. The cog sets the page width (Narrow, Default or Full). Every change saves as you make it. [Project overview](https://docs.lleverage.ai/run-operations/overview/project-dashboard)
* **Point the Agent at what you mean.** The pointer button next to the plus in the Agent's message box starts pointing. Click a row, a cell or a widget on the page, such as an order on hold, and it joins your next message with the value you clicked, so "Why is this one on hold?" needs no more explaining. [Use the Agent](https://docs.lleverage.ai/run-operations/agent/getting-started)
* **Paste a screenshot into the chat.** Paste an image into any Agent message box, such as a supplier's order confirmation, and it attaches with a thumbnail. Text you paste with it stays in the box, so the screenshot and your question go in one message. [Use the Agent](https://docs.lleverage.ai/run-operations/agent/getting-started)
* **Drafts you can copy or send.** An email the Agent writes for you shows as an email card with To, Cc, Subject and **Copy**. A message for a connected Slack or Teams shows in a box naming the channel, and **Send** posts it. A message for anywhere else gets **Copy**, line breaks and bullets included. [Use the Agent](https://docs.lleverage.ai/run-operations/agent/getting-started)
* **New requests, posted in Teams.** In **Organisation Settings**, **Channels**, a pinned Teams channel has a **Post new requests** switch. With it on, every new or reassigned request in the pinned project is posted in that channel, with its title, the project, who it's for and a link to open it. [Set up Microsoft Teams](https://docs.lleverage.ai/administer-and-govern/control/channels/set-up-microsoft-teams)

### Improvements

* **Live widgets show your runs.** A live widget can read the project's own measures: runs, requests and approvals, activity, and tasks automated.
* **Ask your database.** The Agent lists database connections, runs read-only queries on one, and links it to a **Query Database** step.
* **Sort a whole inbox at once.** Gmail **Modify Labels on Messages** labels or archives up to 1,000 emails in one action, so a clean-up needs one approval.
* **Add to a task in progress.** A message you send while the Agent works waits at the end of the chat, with **Send now** and a cross to discard it.
* **Clearer cards in the chat.** A workflow run comes back as one card with how it ended, its first outputs and **Open run**. A finished connection shows the account it connected. Approval cards lead with the question and fold the details under **Details**.
* **A failed save says so.** When a workflow autosave fails, **Couldn't save. Retry** shows beside the workflow's status, and leaving or publishing waits for the save.
* **A new look.** The light theme takes the brand's colours and rounded corners on every page, and the app follows your device's light or dark setting unless you choose one under **Theme**.
* **A fresh session in Teams or Slack.** Send `new session` in your direct chat with the Agent (or `/new` in Teams) to start over. The old session keeps its history in the web app, and **Open previous session** takes you back to it. [The Agent in Microsoft Teams](https://docs.lleverage.ai/run-operations/agent/agent-in-microsoft-teams)
* **Where you are, on every page.** A bar above every page shows the organisation and the project, and switches either.
* **Warnings in the form builder.** A number field's settings have a **Warnings** section, for checks like "the total doesn't match the lines", with a tolerance and a severity.
* **Boards set up by the Agent.** Ask the Agent for a board and it sets up a data table's Board view: the column to group by, card title, sort and lane order.
* **Request emails like the page.** Each request email says what you need to do, shows the request as its page does, and has one button.
* **More.** Form cards in the Agent chat take a file, files in a request keep your zoom, a copy of a default role lists what it can't carry, owners add SSO domains proven by DNS, Dynamics 365 Finance & Operations connects with an app registration, and publishing a workflow names the trigger connection to reconnect.

### Fixes

* Zoho CRM, Zendesk, Slack, Outlook, Excel, Google Drive, OneDrive and SharePoint errors say what went wrong.
* PDFs with more pages than the model accepts are read as text, and encrypted or malformed PDFs are indexed whole.
* Gmail retries after a dropped network connection, and Create Excel accepts sheet names that are numbers.
  {% endupdate %}

{% update date="2026-11-02" tags="new-releases,improvements,fixes" %}

## Week 45: do it once, then hand it over

{% embed url="<https://www.youtube.com/watch?v=BuTb2KV1J54>" %}

This week is about work you only want to explain once. The Agent offers to turn a session into a skill, builds forms from a description and finds records by what they mean. Checking its work gets quicker too: one approval for a batch of calls, a PDF that follows you to a second screen, and file previews on the Files page.

### New features

* **Live widgets: dashboard figures straight from your systems.** A dashboard widget can read a connected system, not only a data table. Ask the Agent for "invoices waiting for approval this month, per approver" and it finds the source, such as Business Central, SAP, AccountView, Gmail or a custom integration like Ridder iQ, previews the figures and puts the widget on your Overview. You can also choose **Live widget** in **Add widget** and tell the Agent what to show. Each widget shows its source and when it last read it, keeps its figures for 5 minutes (15 for slow systems) and reads again when someone opens the page after that, with the last figures on screen in the meantime. A widget only reads, through a connection shared with the project. [Live widgets](https://docs.lleverage.ai/run-operations/overview/live-widgets)
* **Turn a session into a skill.** When the Agent sees that the work in a session is something you'll repeat, such as checking a supplier invoice against its PO and booking it in Exact Online, **Turn into a skill** appears above the composer once the run ends. One click starts the skill build, named after the work. It's offered to builders, once per session. [Skills](https://docs.lleverage.ai/build-and-improve/intelligence/skills)
* **The Agent builds your forms.** Describe a Form or Request Input step in plain words, for example "VAT amount and total follow from the VAT code and net amount, and ask for a reason when the status is On hold". The Agent builds it with calculated fields, rules that show, lock or require a field, line items with record pickers and their filters, and column totals. [Form](https://docs.lleverage.ai/build-and-improve/automate/build-a-workflow/form)
* **Find records by what they mean.** The Agent can search a table by meaning, not only by the exact words. Ask which orders came in with damaged goods and it finds "pallet broken", "dented" and "cracked" in the delivery notes. Add filters to narrow the search. Builders and owners have it in the Agent chat, and it's in the Lleverage MCP. [Tables](https://docs.lleverage.ai/build-and-improve/intelligence/tables)
* **Approve a batch in one go.** When the Agent makes the same approval-gated call several times, such as uploading 16 invoices to Google Drive, one card lists every call with its inputs. **Approve all 16** runs them in order and **Deny all** runs none. In Slack the card has the same two buttons. [Access and approvals](https://docs.lleverage.ai/run-operations/agent/agent-context/access-and-approvals)
* **The PDF follows you to a second screen.** In Requests, **Open in its own window** puts the attachment in a window of its own. Open the next request and the window shows that request's PDF. One window serves every request, and the choice is remembered until you close it. [Requests](https://docs.lleverage.ai/run-operations/request-inbox)
* **See every file before you open it.** The Files page opens in card view, with a preview on every card: an image, the first page of a PDF, a shrunk web page or the first lines of a text or CSV file. A folder opens in place, with a breadcrumb back. Click a file and it opens in a panel beside the list, with **Chat with agent**, **Download** and **View larger**. [Files](https://docs.lleverage.ai/build-and-improve/intelligence/files)

### Improvements

* **Forms point to what to fix.** Submit a form with fields to fix and each one turns red, the cursor lands in the first, and the line by the button counts them: "Fix the 2 highlighted fields to submit."
* **Every file in one viewer.** Every file and attachment opens in the same viewer, so images zoom and rotate like PDFs.
* **A line item's file, one click away.** In a request's line items, each file is a chip that opens the file beside the request.
* **The tab shows what the Agent is doing.** On another tab, the icon moves while the Agent works, pauses when it needs you and shows a check when it's done.
* **Your workspace in your own colours.** The first owner of a new organisation is asked once to theme the workspace, with colours found on the organisation's website or picked by hand. **Fill from website** on the Theme page works for every owner.
* **Retired models leave the pickers.** A step still set to a retired model shows which model it runs on now.
* **One look across the app.** Body text is 14px, buttons are one height with the main action in orange, and tables, error boxes and confirm dialogs each share one design.
* **More on requests and skills.** A request opens at its newest item, a cut-off line item cell shows its full text on hover, skill cards on the Agent home page can be sorted, and skill tests mark the writes they held back as simulated.

### Fixes

* Google Calendar **Get Free/Busy** works, and its errors say what went wrong.
* Agent sessions with more than 32 MB of attachments keep answering.
* The Agent keeps going through every step of a longer task.
  {% endupdate %}

{% update date="2026-10-26" tags="new-releases,improvements,fixes" %}

## Week 44: runs that keep going

{% embed url="<https://www.youtube.com/watch?v=dwDaQDdUay4>" %}

This week is about runs that keep going. A workflow the Agent runs asks you on a card when it needs an answer, the rest of a run can move to someone else in one go, and a run that stops halfway carries on from where it was.

### New features

* **Hand someone the rest of a run.** Reassigning a form step in a workflow run now offers two choices. **Reassign task** moves only this step, as before. **Reassign person** also moves every open step of that run that was with the first person, so the next form, such as **Confirm invoice details** after **Select supplier**, goes to the new person too. The step's header keeps the history of who it was assigned to. [Request input](https://docs.lleverage.ai/build-and-improve/automate/human-handoffs/request-input-page)
* **A workflow needs you? The Agent asks.** When the Agent runs a workflow from the chat and it stops at a Form, Chat, Record Voice or approval step, the chat shows that step as a card above the composer. Fill it in, for example a missing PO number, and the same run carries on to its end. [Use the Agent](https://docs.lleverage.ai/run-operations/agent/getting-started)
* **Carry on a run that stopped.** When an Agent run fails partway, **Continue** picks up where the Agent stopped, with every step it already took in that run. **Resend** still sends your last message again and starts over. [Use the Agent](https://docs.lleverage.ai/run-operations/agent/getting-started)
* **The Agent step saves to Files.** An Agent step in a live workflow can read and save Project Files, the way a **Write Project File** step does. A supplier invoice workflow can, for example, save each invoice check in an `invoice-checks` folder for the team. In Files, the **Source** filter has **From a workflow**. [Files](https://docs.lleverage.ai/build-and-improve/intelligence/files)
* **See what each bar is made of.** A bar chart on a data table can be stacked: choose **Stacked bar**, then the column to **Split by**, and each bar shows a coloured segment per value, such as orders per day by status. Horizontal bars stack the same way, and the Agent builds stacked charts when you ask for them. [Project overview](https://docs.lleverage.ai/run-operations/overview/project-dashboard)
* **Track a date with every run.** A metric's **Value Type** can be **Date**. The **Track Metric** step records a day, such as the delivery date an order confirmation promises, Activity shows it as a date, and Activity and Requests filter on it with a date picker, for example runs whose promised date is before the end of the month. [Metrics](https://docs.lleverage.ai/administer-and-govern/control/metrics)

### Improvements

* **The next step, one click away.** After each answer, up to three suggestions for the next step of the work sit in one row above the composer, such as running a workflow on new emails or connecting a system the work needs.
* **Requests views keep their filters.** Each view keeps your filter edits until you choose **Reset to view**, filters can be shown or hidden per view, and you stay in your view after completing a step. Hide the requests list to give a request the full width.
* **A sidebar grouped by what you do.** The project sidebar groups pages under **Automations** (Workflows, Skills, Channels, Schedules), **Intelligence** (Tables, Memory, Connections, Files, Knowledge), **Monitoring** and **Control**. Members, billing and project settings all open inside **Settings**. [Project navigation](https://docs.lleverage.ai/run-operations/overview/navigation)
* **Connect Notion and BigQuery from the chat.** Ask the Agent to connect Notion or Google BigQuery and it shows a sign-in card instead of asking for a token.
* **Agent steps reach every table in their project.** An Agent step in a workflow can read and write all the project's data tables, the way the workflow's Data Tables steps can.
* **Stop the Agent from Slack.** Say "stop" in the Slack thread and the Agent stops.
* **More on the Agent.** A finished skill build says what the skill still can't do. The Agent can list the organisation's members and invite someone, after an approval card. Item list fields can switch off deleting rows.
* **Dark mode in midnight blue.** The dark theme uses the Lleverage brand's midnight blue.

### Fixes

* Visitors of a public app can cancel their own runs.
* PDF previews load on Safari 17.
* Slack and Teams users without a seat are told they need one, with a link to ask for it.
* After you connect an integration from the chat, you land at the end of the conversation.
* A slow **Search Web** attempt stops after 10 seconds instead of holding up the run.
* SFTP steps no longer fail with an "isDate is not a function" error.
  {% endupdate %}

{% update date="2026-10-19" tags="new-releases,improvements,fixes" %}

## Week 43: the Agent works with your team

{% embed url="<https://www.youtube.com/watch?v=sWEfAG6HfpA>" %}

This week the Agent learned to work with the people around it. It can hand part of a job to a colleague and carry on with their answer, learn a process by watching you do it on a screen call, and it checks with you before it runs or removes anything that needs approval.

### New features

* **Hand a task to a colleague.** Ask the Agent to involve someone, such as "have Mila approve the 8% discount", and it sends them a task. They answer it on a card like a request, or work through it with their own Agent. When they're done, your session picks up their answer and carries on. [Requests](https://docs.lleverage.ai/run-operations/request-inbox)
* **Walk the Agent through your process.** On a voice call, share your screen and show how you handle, for example, an order in Exact Online. The Agent asks as you go and saves each step, screen, field and exception as notes in your Project Files. Then ask it to build a Skill or Workflow from those notes. [Call with Agent](https://docs.lleverage.ai/run-operations/agent/call-with-agent)
* **The Agent asks before it acts.** Before it runs a Workflow that a Skill attaches with **Approval** on, switches a live version off, or deletes a Workflow that is in use, the Agent shows an approval card. For a Workflow in use, the card lists what uses it and what each would lose. [Access and approvals](https://docs.lleverage.ai/run-operations/agent/agent-context/access-and-approvals)
* **The composer finishes your sentence.** After each answer, a suggested next message waits in the composer, and what you type is completed in grey. Press **Tab** to take it. [Use the Agent](https://docs.lleverage.ai/run-operations/agent/getting-started)
* **Copy a skill into another project.** Choose **Copy to project…** in a skill card's menu on the Skills page. The copy lands in the other project as a draft, and the dialog lists what stays behind. [Skills](https://docs.lleverage.ai/build-and-improve/intelligence/skills)
* **Request forms that do the maths.** A form field can follow a formula over other fields, the picked record and line items, and updates while the person fills it in. The value stays editable, and **Use calculated value** puts it back. [Request input](https://docs.lleverage.ai/build-and-improve/automate/human-handoffs/request-input-page)

### Improvements

* **Hand the Agent a whole folder.** **Upload folder** in the composer's plus menu, or drop a folder on the composer, attaches every file in it, named by its path.
* **Skill versions you can open.** In a skill's **Versions** tab, open a version read-only, revert to it, or unpublish the active one. The **Usage** tab lists the sessions that used the skill, and the **Activity** tab lays a build out as one timeline.
* **Files from Slack and Teams.** When the Agent asks for a file in Slack or Teams, attach it to your reply and the run carries on. On the web, the conversation shows as one thread.
* **One note thread per request.** The notes under a request cover its whole Workflow run, so an approver reads the booker's notes and the other way round. [Requests](https://docs.lleverage.ai/run-operations/request-inbox)
* **Dates with a time, and progress bars.** A date column's **Format** menu has **Include time**, and a number column can show as a **Progress bar**.
* **See a step retrying.** On the Workflow canvas, a step with retries shows a **Retrying** pill while they run, and its trace row reads "Retrying (2/3)".
* **More on records and requests.** A card's **Timeline** shows who changed which field, from what to what. Archive several requests at once from the requests inbox. Pick a shared view in the **Table** widget, and download a Files folder as one zip.
* **Run a scheduled Workflow now.** Anyone who can run Workflows can start a scheduled Workflow's live version by hand.

### Fixes

* A cancelled sign-in to an integration shows as cancelled, not connected.
* Filters on Requests metrics include every match, also past 1,000 sessions.
* Gmail **Get Attachment** returns the real file, and Zoho CRM **Get Related Records** works with **Fields**.
* Uploading the wrong kind of file to the Agent's file request is refused with the reason.
  {% endupdate %}

{% update date="2026-10-12" tags="new-releases,improvements,fixes" %}

## Week 42: the Agent builds with you

{% embed url="<https://www.youtube.com/watch?v=FidpYobTwYg>" %}

This week the Agent learned to build most of what you would otherwise build in Lleverage yourself. Ask it to set up a project, a dashboard or a workflow, or talk it through on a call. You can also put it to work inside a workflow. It reads your systems before it asks you anything, and it checks with you before anything goes live.

### New features

* **The Agent sets up your project.** Start from one of six processes, such as Quote & Sell or Pay & Collect, or start blank. The Agent connects your systems, adds the team, writes its instructions and hands over the first jobs, as a plan you can follow. [Projects and solutions](https://docs.lleverage.ai/understand-lleverage/readme/projects-and-solutions)
* **The Agent builds your dashboards.** Ask for what you need to see, such as "orders waiting on a PO and handling time per day", and it adds the widgets to your Overview from your live data. Each role can have its own Overview. [Project overview](https://docs.lleverage.ai/run-operations/overview/project-dashboard)
* **The Agent builds and fixes workflows.** Open the Agent beside the canvas and ask for a change. It knows the workflow you have open, highlights the steps it changes, can run the workflow again after a fix, and asks before it puts anything live. [Build a workflow](https://docs.lleverage.ai/build-and-improve/automate/build-a-workflow)
* **Talk to the Agent while it builds.** Start a voice call and talk a workflow, a dashboard or a whole project into place. The call follows you to any page in the project. [Call with Agent](https://docs.lleverage.ai/run-operations/agent/call-with-agent)
* **Put the Agent to work inside a workflow.** Add the Agent as a step and describe the job in plain words. It can use any Skill or integration you can, and asks a person when something is missing, such as a PO number. [Agent step](https://docs.lleverage.ai/build-and-improve/automate/advanced-workflow-building/agent)
* **Work through records on a board.** Show a data table as a board with a lane per status. Drag a card to the next lane and the record's status changes with it. [Tables](https://docs.lleverage.ai/build-and-improve/intelligence/tables)
* **Skills that improve from how they're used.** A Skill's **Feedback** tab collects every rating on work where it was used. Choose **Improve with agent** and the Agent drafts a change for you to approve. [Skills](https://docs.lleverage.ai/build-and-improve/intelligence/skills)

### Improvements

* **Type while the Agent works.** Send a message mid-task and the Agent takes it into the task it's on.
* **Hand the Agent your files.** Up to 50 files per message, Excel files shown as a real spreadsheet, and Google Drive folders connected to Project Files and kept in sync.
* **Know what the Agent made.** The Agent sidebar lists everything the Agent created in the session, and the Agent links straight to the pages it means.
* **Decide with the full picture.** Zoom into a request's attachments and click through them; in Apps, **Edit and resubmit** corrects a form you already sent.
* **Team and access in one place.** Members and Roles tabs with a simpler role editor, seats and invites as one self-serve flow, and with single sign-on, invites only to your company's domains.
* **Claude Haiku 5.5 for quick steps.** Choose **Fastest** in any model picker.
* **Deleting a Workflow that is in use through the REST API now asks for confirmation.** `DELETE /api/v1/workflows/{workflowId}` answers `409` and deletes nothing when a published version is switched on, the Workflow's schedule is on, a Skill lists it, or a Request is waiting for an answer. The answer lists what uses the Workflow and what deleting it would do to each. Repeat the request with `?confirmInUse=true` to delete anyway; any other value is a `400`. The same applies to `POST /api/v1/workflows/{workflowId}/discard-draft` for a Workflow that was never published. The MCP tools `lleverage_delete-workflow` and `lleverage_discard-workflow-draft` remove nothing from a Workflow that is in use and tell the person to remove it in the app. Integrations that delete Workflows in use without the parameter will now receive `409`. Workflows that nothing uses are deleted as before. [REST API](https://docs.lleverage.ai/build-and-improve/automate/publish-apps-and-apis/rest-api)

### Fixes

* The Requests page opens PDFs reliably after you approve a request, and your queue filters stay in place after a refresh.
* A shared link to a request opens that request, whatever view you are on.
* A voice call stays connected while you sign in to an integration.
* Large table imports that replace all rows finish without timing out.
  {% endupdate %}

{% update date="2026-10-02" tags="new-releases,improvements,fixes" %}

## Week 40: Project Files in workflows and reopened requests

Workflows can now use the same Project Files as people and the Agent, and a completed request can be reopened when something needs another pass. The Agent can also look up who changed what in your organisation.

### New features

* **Project Files in workflows.** Workflow steps can read and write the same Project Files that people and the Agent use, in manual, scheduled and triggered runs. [Files](https://docs.lleverage.ai/build-and-improve/intelligence/files)
* **Reopen a completed request.** Eligible assignees can reopen a completed request and run the work again from that handoff. This creates a new request and replaces pending requests in the same session that the new run supersedes. [Requests](https://docs.lleverage.ai/run-operations/request-inbox)
* **The Agent can check audit history.** Ask the Agent who changed a setting or resource, and when. It only shows what you have access to.
* **GPT-6.1 Sol as an Agent model.** Available in the Agent model list, served through OpenAI's EU endpoint. [AI models](https://docs.lleverage.ai/administer-and-govern/control/ai-models)

### Improvements

* **Switching projects.** Opening and switching projects keeps your browser's Back button working. [Finding your way around a Project](https://docs.lleverage.ai/run-operations/overview/navigation)

### Fixes

* Customer organisations can set up the Microsoft Teams connection themselves again. [Set up Microsoft Teams](https://docs.lleverage.ai/administer-and-govern/control/channels/set-up-microsoft-teams)
  {% endupdate %}

{% update date="2026-09-25" tags="new-releases,improvements" %}

## Week 39: boards, custom roles and calls with the Agent

Data tables can now be worked as boards, projects get custom roles with item-level sharing, and you can talk to the Agent on a live call.

### New features

* **Data tables as boards.** Switch any table to a board with lanes by a text, single-select, yes/no or person field. Drag a card to another lane and the record changes with it. [Tables](https://docs.lleverage.ai/build-and-improve/intelligence/tables)
* **Custom roles and item-level sharing.** Project Admins can create roles with None, View or Edit access per area, and share single Apps, Workflows, Tables and Knowledge Bases without giving access to the whole project. The Members page shows where each person's access comes from, and admins can preview the project as a member. Existing access is unchanged. [Roles and permissions](https://docs.lleverage.ai/administer-and-govern/control/members/roles-and-permissions)
* **Call with Agent.** Start a live voice call with the Agent when talking is easier than typing. The Agent can also suggest a call when it needs you; the call starts when you press **Call**. [Call with Agent](https://docs.lleverage.ai/run-operations/agent/call-with-agent)
* **Overviews that build themselves.** An Overview you haven't customised grows with the project: Apps, Workflows, Skills, Tables, open requests and recent items, plus a **How it is going** section once a project has 100 runs in 30 days. Editing turns it into a custom dashboard; **Clear** goes back. [Project overview](https://docs.lleverage.ai/run-operations/overview/project-dashboard)
* **Feedback on Agent work.** Rate a single Agent response or a whole session, with an optional comment, and change it later. Builders see the feedback next to the run in Activity.
* **Skill Tests for builders.** Builders with the right project access can run tests from a Skill or a Skill proposal. Tests never publish or change the live Skill on their own. [Skills](https://docs.lleverage.ai/build-and-improve/intelligence/skills)

### Improvements

* **SharePoint document libraries and large lists.** The Agent can find a site's document libraries and read their files and folders, and large lists are read page by page. Microsoft's limits still apply: a read stops at 5,000 items or 50 pages. [SharePoint](https://docs.lleverage.ai/administer-and-govern/control/connections/sharepoint)
* **Replying while an approval is pending.** You can keep typing while the Agent waits for you to confirm a protected change. Your reply adds context; only the confirm button approves the change.
* **Built-in industry knowledge.** The Agent can use general industry and process knowledge to ask better questions during setup. Your own rules, policies and master data still come from your project and systems.
* **Dashboards in the Lleverage style.** Dashboards the Agent builds use the same look as Lleverage Overview pages, with your organisation's theme.
  {% endupdate %}

{% update date="2026-09-18" tags="new-releases,improvements,fixes" %}

## Week 38: clearer failed runs and a new navigation

Failed workflow runs now explain what happened and who can fix it, and the navigation is cleaner and adjustable.

![Widget settings open in the right sidebar](https://3370306531-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FlgSglJDQZxDZdn20CUcl%2Fuploads%2FHFxsd8y9VnMjEGoKPLlh%2Fwidget-right-sidebar-edit-v2.png?alt=media)

### New features

* **Failed runs explained in plain language.** A failed run opens with **What happened** and **What you can do**, in your browser's language where possible, with the raw error underneath. Anonymous Workflow-app viewers still see the raw error.
* **Why a run failed, and who acts next.** Failures are classified, for example bad input, a workflow or script error, a provider outage or a timeout, and point to the person who can act next. When the evidence isn't clear, the cause stays **unknown**. Project Admins and Builders get an in-app notification for production failures. [Monitoring](https://docs.lleverage.ai/build-and-improve/monitoring)
* **Richer charts on Monitoring dashboards.** Charts built on data tables can be split into series, with your own labels and colours, a date column bucketed by day or week, thresholds and comparisons with the previous period. [Monitoring](https://docs.lleverage.ai/build-and-improve/monitoring)
* **Edit widgets in a side panel.** On Overview, Monitoring and Agent pages, switch to edit mode, pick a widget and set it up in the panel on the right. [Project overview](https://docs.lleverage.ai/run-operations/overview/project-dashboard)

### Improvements

* **A cleaner, adjustable navigation.** Collapse the sidebar to icons or drag it to the width you want. Pinned pages show recent items on hover, and your organisation's highlight colour runs through the app. Pages stay where they were. [Finding your way around a Project](https://docs.lleverage.ai/run-operations/overview/navigation)
* **Messages and tool calls counted apart.** An Agent session's Activity counts messages and tool calls separately, so it matches the transcript.

### Fixes

* An Agent step in a workflow that returns no usable text, files or structured result now fails, instead of passing an empty value to the next step. The same goes for missing or malformed JSON output. [Agent step](https://docs.lleverage.ai/build-and-improve/automate/advanced-workflow-building/agent)
  {% endupdate %}

{% update date="2026-09-12" tags="new-releases,improvements,fixes" %}

## Week 37: usage-based billing and Microsoft Teams setup

Usage-based billing is live where it is enabled, customers can connect Microsoft Teams themselves, and requests get easier to check against their source PDF.

### New features

* **Usage-based billing.** Agent seats are separate from organisation membership. Owners set monthly Agent-usage limits per organisation, project and member, and members can ask for a seat or a higher limit. Reaching a limit pauses new Agent messages, not workflows or scheduled runs. Lleverage turns this on per organisation. [Subscription management](https://docs.lleverage.ai/administer-and-govern/control/settings/subscription-management)
* **Set up Microsoft Teams yourself.** Organisation Owners can start the Teams connection in Lleverage and finish it with their Teams administrator. [Set up Microsoft Teams](https://docs.lleverage.ai/administer-and-govern/control/channels/set-up-microsoft-teams)
* **Teams channel messages with images and files.** The Teams channel-message trigger passes images and supported shared files into the workflow.
* **Check a request against its PDF.** Hover over a field in a request and the matching value lights up in the PDF beside it. Works on searchable PDFs, up to the first 100 pages. [Requests](https://docs.lleverage.ai/run-operations/request-inbox)
* **Edit line items cell by cell.** In Request Input and Request Decision forms, click a cell in an item list to change it. **Enter** saves, **Escape** cancels. [Request Input](https://docs.lleverage.ai/build-and-improve/automate/human-handoffs/request-input-page)
* **Retry a failed Agent response.** A failed Agent response has a **Retry** button that runs your last message again.

### Improvements

* **Agent reports in your style.** HTML reports the Agent makes use the Lleverage look, and your organisation's branding when it is set.
* **Faster and richer Agent HTML.** Short summaries show straight in the conversation; richer reports with tabs open as a file in a preview.
* **Teams conversations reach project connections.** An Agent conversation in Teams that isn't pinned to a project can use connections from the projects you have access to. [Agent in Microsoft Teams](https://docs.lleverage.ai/run-operations/agent/agent-in-microsoft-teams)
* **Project API tokens follow their creator.** A Project API token stops working when the person who created it leaves or loses access to the project. **Action needed:** replace tokens owned by someone who is leaving before they go. [API tokens](https://docs.lleverage.ai/administer-and-govern/control/api-tokens)
* **API access is per project.** Organisation-wide API tokens are removed. **Action needed:** integrations that used one need a Project API token. [API tokens](https://docs.lleverage.ai/administer-and-govern/control/api-tokens)

### Fixes

* Files uploaded in a one-to-one Teams conversation, including inline images, reach the Agent.
  {% endupdate %}
  {% endupdates %}


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.lleverage.ai/changelog/readme.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
