Who this is for. AP Helpdesk Analysts. See the Roles and Access.
Up to six panels attach to the task detail pane. Three open from links at the top right of the pane: AI Reasoning and Related Info, What the agent did and Activities. Task Notes and Records are always tabs down the right edge; Email Thread joins them only on a task that has an email thread behind it.
Figure 1. The right-edge tab rail.
Table 1. Panels and where they open from
| Panel | Opens from | Contains |
|---|---|---|
| AI Reasoning and Related Info | Link at the top right of the detail pane | A stage progress indicator for the pipeline the task ran, and the draft. |
| Activities | Link at the top right of the detail pane | A timeline headed Task Activities. |
What the agent did |
Link at the top right of the detail pane |
What was decided, who decided it, and what the agent considered and passed over. Shown on a task that ran a pipeline. |
| Email Thread | Tab on the right edge, on a task that has an email thread | The messages on the task. |
| Task Notes | Tab on the right edge | A note editor and note list. |
| Records | Tab on the right edge | Six tabs of vendor records. |
AI Reasoning and Related Info and Activities open as overlay drawers with a close control. The three edge tabs open in the right of the pane.
One panel at a time
Only one edge panel can be open at once. Opening Email Thread replaces the rail: the Task Notes and Records tabs are withdrawn while it is open, and return when it is closed. The same holds for each of the three.
Panel state does not follow the task. Selecting a different task in the queue closes whichever panel was open, and the new task opens with the pane in its default state. A panel you want to consult on several tasks has to be reopened on each of them.
Which panel answers which question
Table 2. Choosing a panel
| You want to know | Open |
|---|---|
| Why the agent reached this conclusion | AI Reasoning and Related Info |
| Who changed what, and when | Activities |
| What the vendor actually wrote | Email Thread |
| What a colleague already tried | Task Notes |
| What the underlying bill or payment says | Records |
AI Reasoning and Related Info
Selecting AI Reasoning and Related Info at the top right of the task detail pane opens an overlay drawer over the right of the screen, with a close control at its top right. The drawer holds a row of stage tabs for the pipeline the Agent Co-worker ran and, beneath them, the panel for the stage you have selected. It opens on the stage the pipeline has reached, which for a task waiting on your review is the draft reply. Select any other stage to see its detail in the same place: the triage decision, the detected email categories, or the system-of-record (SOR) matching.
Stage indicator
The stage bar follows the pipeline, so what you see depends on what the Agent Co-worker did. A task with a reply shows four stages: Triage, Email Category Detection, SOR, and Draft Response, joined by connecting lines. A task that submits an invoice or credit memo adds Submission as the fourth stage, with Draft Response fifth. A task the Agent Co-worker escalated or forwarded shows Escalate in place of Draft Response, relabeled Escalated to <assignee> or Forwarded to <team> once it completes. On a task held for policy review, the third stage is labeled Policy SOR.
Table 3. How a stage is shown
| State | Shown as |
|---|---|
| Completed | A filled circle carrying a check mark. |
| Selected | A filled circle carrying the stage number, for example 4 on Draft Response. The number marks the stage you are viewing, and the drawer opens with the stage the pipeline reached already selected. |
| Not reached, or not selected | The stage’s own icon in an outlined gray circle. |
| Skipped | Grayed, with the label struck through. A stage being reprocessed after a rewind is grayed without the strikethrough. |
Figure 2. The AI Reasoning drawer: the stage indicator, with the reply block beneath it.
The draft in the drawer
Below the stage indicator the drawer carries the whole reply block, not a preview of it: From, To with Cc and Bcc, Subject, the draft with its language selector and Original / Translated toggle, the formatting toolbar, Attachments, Tone, and the Approve & Send, Discard and Forward actions. A saved indicator sits beside the Draft label.
It is the same draft, not a copy. The draft here is fully editable, and it is the same one as in the main pane: text typed into the drawer appears in the main pane editor immediately, and the two always show identical content. Work in whichever is convenient. There is no second draft to reconcile, and no need to re-enter anything.
Because the drawer is narrower than the main pane, the later toolbar controls are folded under the overflow control at the end of the toolbar rather than being shown inline.
Activities
Selecting Activities at the top right of the task detail pane opens a drawer headed Task Activities, holding a timeline of changes to the task, newest first.
Figure 3. The Task Activities timeline.
What an entry contains
Every timeline entry breaks into the same parts, so you can tell who acted, what changed, and when, at a glance.
Table 4. Activity entry
| Part | Example |
|---|---|
| Timestamp | August 18, 2026 at 08:28 PM IST |
| Actor | Agent via [DP-9/DR-4], or a person’s name |
| Change | changed Assignee |
| Before and after | Unassigned > Jordan Ellis |
What is recorded
Both agent actions and actions taken by a person are recorded, and both use the same layout: timestamp, actor, change, then the before and after values where the change has them. Only the actor differs.
Table 5. Entry types
| Entry | Actor |
|---|---|
| created the Task, with Initial: status New, priority Medium | Agent |
| changed Status, for example New > Open | Agent via [DP-9/DR-4] |
| changed Assignee, for example Unassigned > Jordan Ellis | Agent via [DP-9/DR-4] |
| updated Draft tone, for example Balanced > Concise | The person who changed it |
| regenerated Draft | The person who regenerated it |
| answered Gap questions | The person who answered them |
| edited Draft | The person who edited it |
Retoning, regenerating, answering gap questions, and hand editing are each recorded separately.
The actor field
Agent entries carry a bracketed policy and rule identifier after the actor, in the form Agent via [DP-9/DR-4], when a distribution policy or rule drove the change. Where no policy or rule applies, the actor reads Agent. An assignment held to the current assignee reads Agent via Assignee Hold, and an assignment made by an NLU rule reads Agent via NLU rule. Entries made by a person carry the person’s name with no bracketed identifier.
Timestamps
Timestamps follow the time zone of the person reading them, taken from the browser, not the tenant and not a fixed setting. The same entry that reads August 19, 2026 at 08:01 AM IST for a reader in India reads August 18, 2026 at 10:31 PM EDT for a reader in New York. It is the same instant, relabeled. An AP team and its administrators in different regions therefore each see their own local time, and neither has to convert.
Every entry carries a full timestamp with an explicit time zone abbreviation. The same format and zone are used by the timestamps on Task Notes.
What the agent did
What the agent did opens from a link at the top right of the detail pane, beside AI Reasoning and Related Info. Where the reasoning panel reports what each stage of the pipeline produced, this one reports what was decided, who decided it, and what the agent considered and passed over. It appears on a task that ran a pipeline.
Email Thread
The Email Thread tab on the right edge of the task detail pane holds the messages on the task.
Figure 4. The Email Thread panel.
Email Thread controls
Alongside the messages, the panel gives you a count, a sort order, and a way to reply from inside the task.
Table 6. Email Thread controls
| Control | Shows |
|---|---|
| Message count | The number of messages, for example “1 message”. |
| Newest first | A sort control at the top right. |
| Reply | A button at the foot of the thread. |
What each message shows
Every message carries an avatar, the sender’s name and email address, a to: line, and a date and time. External senders carry the External Sender badge.
Each message header also carries its own language selector showing EN, which translates that message into the language you pick. A two-state toggle appears beside it only when automatic pre-translation is switched on in your user preferences, or once a translation has been produced for the message. The states read English (Original) and English (Unavailable), the second dimmed, because the inbound message was already in English, so no translation exists to show.
Show message metadata, an icon in the message header, does not open anything. Selecting it replaces the detail pane with an error, and the pane recovers on a page reload.
Reply from the thread
Reply opens a reply composer inside the Email Thread panel. It does not scroll to the draft reply in the main pane, and it does not alter that draft. It is a separate, blank reply.
The composer carries a recipient chip, an Add Cc / Bcc control, a rich text editor, Attach local files, and its own Send button. The panel stays open while you compose.
Figure 5. The reply composer inside the Email Thread panel.
There are two ways to reply to a vendor, and they are not the same thing. The draft reply in the main pane is the one the Agent Co-worker prepared, with Tone, translation, generated attachments, and the Approve & Send / Discard / Forward actions. Reply here composes a fresh message by hand. See Review and Send the Draft Reply.
Forwarded messages
A message in the thread can be a forward, carrying the original sender, date, subject, and recipient inside its body below a “Forwarded message” separator. The same content can appear twice in one thread: once forwarded into the mailbox, and once received directly from the external sender.
Task Notes
The Task Notes tab on the right edge of the task detail pane holds notes recorded against the task.
Figure 6. The Task Notes panel with two notes saved.
Task Notes controls
Table 7. Task Notes controls
| Control | Shows |
|---|---|
| Search notes | A search field at the top of the panel. It is always present, including on a task with no notes. It matches on note text and author name, and it also searches replies. |
| Formatting toolbar | Bold, Italic, Underline, Bullet list, Numbered list, Blockquote, and Source code. |
| Character counter | Reads 0/2000 with the editor empty. |
| Add note | Unavailable while the editor is empty; becomes available as soon as text is entered. |
A task with no notes reads “No notes yet.”
The character limit is per note
The counter tracks the note being composed, not the task. Typing an eighty-four character note moves it to 84/2000, and saving the note returns it to 0/2000 ready for the next one. The 2,000-character limit therefore applies per note, and a task can carry any number of notes.
Order and sorting
Notes are listed newest first. A sort control sits to the right of the search field: select it to switch the list between newest first and oldest first. The icon flips to show which order is in effect.
What a saved note shows
Each saved note carries the author’s initials as an avatar, their name, their email address in angle brackets, and a timestamp with an explicit time zone, above the note text.
Note actions
Hover over a saved note to reveal its actions. Reply appears on every top-level note. Edit and Delete appear only on notes you wrote yourself, and not on notes written by the AI Agent or on notes already deleted. Delete asks you to confirm before it removes the note, and a note you have edited is marked (edited) beside its timestamp.
Agent notes and user notes
The panel holds two kinds of note, and they behave differently. User notes are the ones people write. Agent notes are written by the Agent Co-worker as it works: a timestamp, what the sender asked, which records it could or could not match, the confidence it had, and a link back to the message. The edit and delete controls appear only on your own user notes. Agent notes cannot be edited or removed, which is what makes them usable as a record of what the agent did.
The same Agent / User split appears on vendor, bill and payment records, where each side carries its own count. If a task looks unexplained, the agent notes on the matched vendor record often carry the reasoning the task does not.
Notes are not part of the task activity timeline. Adding a note records nothing in Task Activities, which tracks changes to the task and its draft only. If you need a decision to appear in that timeline, a note alone will not put it there.
Records
The Records tab on the right edge of the task detail pane opens a browser over the records held for the vendor the task was matched to. The panel is headed Matched Vendor: when the sender is external and Linked Vendor: otherwise, followed by the vendor name, and everything in it is scoped to that vendor. Where the task matched more than one vendor the name is a picker you can switch between; where it matched none, the panel shows All vendors, No vendor matched, or No vendor linked in place of a name.
Figure 7. The Records panel, Snapshot tab.
Tabs
Six tabs run across the top of the panel: Snapshot, Invoices, Payments, POs, Credits, and Vendor. A Search field sits alongside them.
Snapshot holds three blocks:
IDENTITY, holding Vendor, Vendor ID, and Email.
RECENT COMMUNICATION, a list of messages showing subject, sender, and date.
QUICK JUMP, holding links labeled Invoices, Payments, Purchase Orders, Credit Memos, and Vendor master.
Invoices, Payments, POs, and Credits
Each record tab opens with a count and the sort order, in the form “43 invoices · Sorted by date descending”. Every row carries the record reference, a status, the date, the vendor, and an amount or currency; an amount that is not held shows as an em dash.
Longer lists paginate. The Invoices tab may read “Page 1 of 5 · 1–10 of 43” with ‹ Prev and Next › controls; shorter lists such as POs and Credits show every row without pagination.
Figure 8. The Invoices tab.
Table 8. Statuses by record type
| Tab | Statuses seen |
|---|---|
| Invoices | OPEN, PAID, CANCELED |
| POs | OPEN, CLOSED |
| Credits | OPEN |
Which statuses appear depends on the connected system of record.
Figure 9. The Credits tab.
Vendor
The Vendor tab holds the vendor master record in two blocks. FIELDS carries Vendor Name, Entity, Contacts, ID, Email, Business Name, Category, Tax ID, Tax ID Type, Status, and Address. Beneath it a second block is named after the connected system of record and carries the fields specific to it, such as Supplier Group and Default Payment Type. Values not held show as an em dash.
Figure 10. The Vendor tab.
That the second block is titled with the name of the connected system of record, and changes with it, indicates these values are passed through from the system of record rather than being Auditoria’s own. Treat the panel as a read-only window onto the ERP.
The search field
The search field filters within the tab you are on. Searching a purchase order number while the Invoices tab is open returns “0 invoices” and the empty state “No invoices on file for the current scope”. It does not cross to the POs tab and does not leave the matched vendor.
When the task matched a vendor, the panel is locked to that vendor. If the Agent Co-worker matched the wrong vendor, the records shown are the wrong vendor’s, and searching here will not reach the right ones. When the task matched no vendor and the sender is internal or system, the panel opens in all-vendor scope instead: the Vendor tab lists every vendor, choosing one scopes the other tabs to it, and a Back to all vendors link clears the choice.
Related information
For the pane these panels attach to, see Read a Task.
For the draft the AI Reasoning drawer also carries, see Review and Send the Draft Reply.