The screening agent page

Everything one screening agent produces and everything that governs it — the reports from each run on one side, and on the other the portfolio it screens, the concepts it looks for and the methods it uses.

Opening the page

  1. Click Agent Marketplace in the left rail.
  2. Under My Agents, click Screening Agents.
  3. Click the agent's card.
Other ways in

When a test run finishes, the Batch test run completed dialog offers View Results, which opens the agent here. Creating a screening agent from a template also ends on this page.

Which page you land on

An agent that has already produced reports opens on Results & reports. One that has not opens on General. Once you have chosen a page yourself, the page does not move you again.

A screening agent open on Results & reports. The page title is the agent's name. A narrow menu at the left has a RESULTS group with Results & reports expanded to a report and two dated runs, a CONFIGURATION group with General, Concepts and Methods, and an ESTIMATED COST card at the bottom. The main area has a bar with Frequency and Last Run on the left and HTML and Markdown buttons, a share icon and a download icon on the right. Below it the report fills the page: a title and summary lines, filters for Company, Industry and Country with a Columns list and a blue Create portfolio button, tabs for Companies, Explore, Portfolios and Changes, a legend of sources, and a table of companies.
There are no Test Run, Publish or delete buttons in the header on this page.

How the page is laid out

A secondary menu lists everything about this agent, split into two groups. The button at the top left of the main header collapses and restores it. The agent's name replaces the generic title in the page's breadcrumb.

GroupEntries
RESULTS Results & reports — what the agent has produced. It expands to a tree of reports.
CONFIGURATION General, Concepts, Methods — what it will do next time it runs.
Not the same as a custom agent

A screening agent has no Structured Data, Run History, Data Source, Data Builder or Report Builder pages. Its scope is a portfolio, its criteria are concepts, and its output is the reports. Compare The custom agent page.

Estimated cost

ESTIMATED COST sits at the bottom of the menu and stays there whichever page you are on. It gives the credits one Run Once of this agent would consume at its current configuration, and updates as you change the portfolio, the concepts or the methods.

ControlWhat it shows
Round exclamation mark beside the figure Opens Cost breakdown, headed by the number of companies and concepts. Companies, Concepts and Methods each expand: Companies to the portfolio and its company count, Concepts to the credits each concept costs, Methods to each method's rate (Off for a method switched off). A formula line and a Total Credits figure close the panel. Not available while the page is still loading.
View details Opens Credit, where your balance and past usage live.

The estimate is the number of companies in the portfolio × the number of concepts × the sum of the rates of the methods that are switched on. Each method has its own rate per company per concept, except Data Feed - Bloomberg, which has none.

The four rates are listed on the creation page. In the picture below every method is switched on, so each company and concept pair costs the sum of the rates; Data Feed - Bloomberg is listed with 0.00 credits because it adds nothing. A method that is switched off reads Off instead.

The Cost breakdown panel open over the left of the page, with the report behind it. It opens with a heading giving the number of companies and concepts. Three rows are expanded: Companies lists the portfolio and its company count; Concepts lists each concept with its credits; Methods lists the five methods, each with a rate, the last one, Data Feed - Bloomberg, reading 0.00 credits. A line writes out companies × concepts × rate, and a Total Credits bar closes the panel.
The panel is anchored beside the ESTIMATED COST card, so it covers the menu and part of the page.

Header actions

They are not always there

Test Run, Publish and the delete button appear in the main header only while you are on one of the three CONFIGURATION pages. Switch to Results & reports and they go away.

Test Run

Runs the agent once, now, against its current configuration.

  1. Click Test Run to open the panel.
  2. Leave Run Now selected. The panel spells out the dates it covers: Structured data from data source: one year ago through today.
  3. Click Run. If the agent has no email recipient, a prompt asks about email first — see Email prompt before a run.
The General page with the Test Run panel open under the header. The panel has a Run Now option with a radio mark, a line beginning Structured data from data source with a start and an end date, and a dark blue Run button at its bottom right. Test Run, Publish and a red bin icon are in the header above it.
Run was not clicked for this picture. The dates run from one year ago to today.
A test run spends credits

The cost is the ESTIMATED COST figure, checked for the methods that are switched on. If your balance will not cover it, the run is refused and a dialog explains why. Data Feed - Bloomberg has no credit rate in the credit check, so switching it on adds nothing to the figure.

Email prompt before a run

Clicking Run on an agent whose Email card has no recipient does not start the run straight away. A prompt titled Email notification is off reads Turn on email notification to get the report by email when this run finishes. and offers three ways out:

ChoiceWhat happens
Turn on email Opens Edit email with your own address already filled in, when Orbit knows it. Once you save there, the run starts. If you close that dialog without saving, nothing runs.
Run without email Starts the run now, with email left off.
Close (the × or Esc) Cancels. The run does not start.

The prompt does not appear when the agent already has at least one address on its Email card.

A small dialog titled Email notification is off over the dimmed General page. It has an information icon, the sentence Turn on email notification to get the report by email when this run finishes, a close button at the top right, and two buttons: Run without email and a dark blue Turn on email.
It appears after Run is clicked in the Test Run panel, and only when the agent has no email address. Neither button was clicked for this picture.

Once a run starts, progress is shown as Test run in progress with a percentage, in a bar at the top of the window that survives navigating elsewhere. When it finishes, the Batch test run completed dialog appears — Click below to view your results, or continue with your current workflow. View Results opens this page again; Cancel dismisses it.

No scheduled date option

Unlike a custom agent, a screening agent's Test Run offers only Run Now; there is no Run at specific time. Edits you have made are already saved, so what runs is always what the configuration pages show.

Publish and Move to draft

A new agent is a Draft: it exists and can be test-run, but its scheduled tasks do not fire. Publish activates the agent and enables all of them. Once published the button reads Move to draft and does the reverse. Hovering it explains which: Activate agent and enable all scheduled tasks. or Deactivate agent and disable all scheduled tasks.

ButtonBehaviour
PublishActivates immediately — no confirmation — and confirms with Now published.
Move to draft Asks first: Move this agent to draft? Users won't be able to access it until republished. Draft confirms, Cancel backs out. Confirming shows Moved to draft.

Deleting the agent

The delete button carries a bin icon rather than a label — hover it to see Delete.

This cannot be undone

The confirmation is headed Are you sure you want to delete this agent? and reads Once deleted, any structured data and reports produced by this agent will no longer be available. Confirming with Yes, Delete removes the agent and returns you to Agent Marketplace.

There is no Save button

Every change on the configuration pages is saved the moment you confirm it — a dialog's save button, or a switch being flipped. A Saved notification confirms each one; a Failed notification means the change did not go through.

Results & reports

The documents the agent writes. Each run of the agent produces a new report, so this page is a two-level list before it is a reading view.

Finding a report

  1. Click Results & reports in the menu. Its report tree expands underneath.
  2. Click a report to list its runs, newest first. Clicking it again collapses it. The number beside it is how many runs it has.
  3. Click a run to read it. The list holds the latest 50 and is headed Latest 50 reports.
What opens by default

The first report that has any runs opens on its own, showing the most recent one. Reports with no runs yet stay collapsed.

Reading a report

The bar above the report belongs to Orbit; everything below it is the report itself.

ItemMeaning
FrequencyHow often this report runs. See Schedule.
Last RunThe date the report was last produced, as YYYY-MM-DD; - when unknown.
HTML / Markdown Switches how the report is rendered; the selected one is filled dark blue. A format the run did not produce is greyed out, as are both while one is loading. A Markdown view with no content shows a no-data placeholder.
Share icon A three-dot share icon. Hover it for Share. Creates a link to this report and copies it to your clipboard, confirming with Copied to clipboard. The link is then listed on Share.
Download icon A downward arrow into a tray. Saves the report as a file named after the report — .html or .md according to which format is showing. It has no tooltip.
The same report with Markdown selected in the bar, so that button is filled dark blue. The report is now shown as plain text on a light background: a heading, bullet points and then the filters and table contents as separate lines of text, without the filter controls or table layout.
Markdown shows the report's text content only.
Before any report exists

The page reads No report yet — This agent has no reports yet. Once it runs, the generated report will show up here.

Saving the result as a portfolio

A report can hand the companies it found back to Orbit so you can keep them as a portfolio. That control lives inside the report itself, not in Orbit's own bar — the report's filter bar carries a Create portfolio button. When the report asks Orbit to create a portfolio, Save result opens and the report waits for the outcome.

Two different jobs

Save result only creates a new portfolio. Adding companies to a portfolio you already have is done inside the report, where you confirm a preview of the change; Orbit then applies it without opening this dialog. A report that has already updated a portfolio cannot do it a second time.

ItemNotes
Intro line These N companies are saved as a result with an AI-written report. The list itself goes into a portfolio.
Portfolio name Required; placeholder Enter portfolio name. Save and write the report stays disabled while it is blank.
Note under the box A new portfolio is created in Portfolio management holding these N companies. The portfolio appears in Portfolio Management.
Save and write the report Creates the portfolio and closes the dialog. Orbit shows no Saved notification of its own here — the report confirms the result itself.
Cancel (or the × at the top right) Closes the dialog without creating anything; the report is told the request was cancelled.

If creating the portfolio fails, a Failed notification shows the server's reason, or Failed to save result when it gives none. The dialog stays open so you can change the name and try again. A failed update of an existing portfolio shows Failed with Portfolio "<name>" not updated: followed by the reason.

General

The agent's identity, who it emails, the portfolio it screens and when it runs — plus the two ways of reaching it from outside Orbit. Each card except the last two has a pencil icon that opens its own edit dialog.

The General page. The header holds Test Run, Publish and a red bin icon. A stack of cards runs down the page, each with a heading and, for the first four, a pencil icon at the right: the agent card with its name, a green status badge and a description; Email, reading No email configured; Portfolio with a portfolio name and a company count at the right; and Schedule with Frequency, Day and Time on one line and Data range and Last run on a second. Below are MCP/API Use, with an off switch, a link and a dimmed endpoint box with a disabled Copy button, and Chat Use, with an off switch and a sentence about it.
The agent id in the endpoint is an example. The endpoint box is dimmed while MCP/API Use is off. In the Schedule card, Data range says how far back each run looks, and Last run gives the dates the most recent run covered.

Agent information

The card carries the agent's name, a status badge, and its description. The pencil at its top right opens Edit basics.

FieldNotes
Agent nameRequired. Save changes does nothing while it is blank. An agent saved without one reads Untitled Agent.
DescriptionFree text, two lines tall. Shown on the marketplace card. Reads -- when empty.

Save changes applies the edit and confirms with Saved; Cancel discards it.

The Edit basics dialog over the dimmed General page: an Agent name box, a Description box, a close button at the top right, and Cancel and a dark blue Save changes button.
Agent name is required; Save changes does nothing while it is blank.

Email

Who receives the agent's email. The card lists the addresses separated by commas, or reads No email configured. The pencil opens Edit email.

FieldNotes
Email A multi-select you can type into to filter. The choices are the email addresses of the members of your organisation (see Organization); if there are none to list, your own address is the only choice. Leaving it empty turns email off. Many selections collapse into a counter with a hover list.

Save changes / Cancel, as above.

The Edit email dialog: an Email multi-select showing its Email placeholder with a drop-down arrow at its right, above Cancel and Save changes.
Leaving the box empty turns email off. Addresses you pick appear in the box as removable chips.

Portfolio

which companies this run covers — the card shows the portfolio's name and its company count (No portfolio selected when there is none). The pencil opens Edit portfolio.

ControlNotes
Portfolio Opens a list of your portfolios, each with its company count, with a Search portfolio box. Reads No portfolios found when the search matches nothing. The current portfolio is pre-selected.
Portfolio management Link in the note under the field: Portfolios are created and maintained in Portfolio Management. Leaves this page.
ApplySaves the choice and confirms with Saved — Portfolio has been updated. Does nothing until a portfolio is selected. The cost estimate updates to the new company count.
CancelCloses the dialog.
The Edit portfolio dialog: a Portfolio drop-down showing the current portfolio, an information line saying portfolios are created and maintained in Portfolio management, with a link, and Cancel and a dark blue Apply button.
Apply does nothing until a portfolio is selected.

Schedule

How often the published agent runs, and how far back each run looks. The card shows Frequency, the Day where it applies (a weekday for Weekly, a day of the month for Monthly, Quarterly and Yearly), and Time. Under them a second line shows the time window. The pencil opens Edit Schedule.

On the cardMeaning
Data range The time window in words, followed by (each run covers the data before its run date). 1M reads Last 1 month, 3M Last 3 months, 6M Last 6 months, YTD Year to date and 1Y Last 1 year. The line is left out when the agent has no window stored.
Last run The dates the agent's most recent report actually covered, as YYYY-MM-DD – YYYY-MM-DD. Left out until the agent has a run with both dates.
FrequencyWhen it runs
DailyEvery day at 8:00 AM.
WeeklyEvery Monday at 7:00 AM.
MonthlyThe 1st of each month at 8:00 AM.
QuarterlyJan 1, Apr 1, Jul 1 and Oct 1 at 6:00 AM.
YearlyJanuary 1st at 5:00 AM.

The dialog has two lists. Frequency starts on the agent's current frequency, with the line above for the option you pick underneath it. Time window comes next and is shared by every method on the Methods page. It starts on the agent's stored window, or 1Y when it has none. Save Change saves both and confirms with Saved — Schedule has been updated. Cancel discards them. The schedule only fires while the agent is published.

Time window optionNotes
1M, 3M, 6M, YTD, 1Y Under the list: Every filing type is searched within this window — annual, interim and quarterly reports, earnings call transcripts, ESG reports, AGM materials, company announcements and IR presentations.
Two places show a time

The line in the dialog is the standard time for the frequency you pick. The Time on the Schedule card is read from the schedule stored on the agent, so the two can differ. For the agent shown on this page they agree: the card reads 6AM and the dialog says 6:00 AM. When they disagree, the card is what is stored on the agent.

The Edit Schedule dialog: a Frequency drop-down with a small grey line beneath it giving the run dates and time for that frequency; a Time window drop-down with a grey note about which filing types are searched; and Cancel and a dark blue Save Change button. The Schedule card behind the dialog shows Frequency, Day and Time.
Save Change here is singular, unlike the Save changes buttons in the other dialogs. The time window applies to every method.

MCP/API Use

Exposes this agent over MCP so tools outside Orbit can call it. The switch is off by default. Turning it on saves at once, marks the card Protocol Ready and enables the endpoint below.

ControlNotes
MCP endpoint The address to paste into Claude Desktop, Cursor or another MCP client. It ends in the agent's own id, so every agent has its own. Dimmed, with Turn on MCP/API use to enable., while the switch is off.
Copy Copies the endpoint and confirms with Copied — Copied to clipboard. Disabled while MCP/API Use is off.
Get started Opens Orbit's developer documentation for Agent Builder in a new browser tab.
See also

MCP covers connecting Orbit to an MCP client in general; this switch is only about exposing this one agent.

Chat Use

Lets you put questions to this agent's data in conversation. Also off by default; flipping it saves at once. Either switch that fails to save returns to where it was and shows a Failed notification.

Concepts

The concepts the agent screens each company for. A concept is a reusable definition of one thing to look for — see Concept Library.

The Concepts page: a Concepts card with a pencil icon at the top right. Each concept is one row with an arrow at the left, its theme in small grey text, its name in bold, and a count of keywords and extraction fields at the right. The first row also has a pencil icon at its far right; the second does not.
The pencil beside the heading changes which concepts the agent uses. The pencil on a row edits that concept and is missing on public concepts such as the second one.

Reading the list

One row per concept: an arrow, its theme in small text, its name in bold and, at the right, a count in the form N Keyword · M extraction field followed by a pencil icon. With none added, the list reads No concept selected. Click the arrow, or the name, to expand a row.

Expanded sectionWhat it holds
PDF KEYWORDS The keyword groups the concept searches filings for, each with its keywords as chips, and the total count.
EXTRACTION FIELD n One section per field: its name, a badge for its type (Number, with the unit when there is one; Yes / no; Category; or, for a field that takes its value from Bloomberg, Bloomberg · <mnemonic>), and the extraction rule in words. A Bloomberg field is scored with the same kind of rules as a number, yes / no or category field.
BLOOMBERG FIELD Shows Key followed by the Bloomberg field identifier the concept uses, or — when it uses none. A concept can use only one Bloomberg field. This section appears only when Bloomberg is set up for you — see When Bloomberg appears.
Score preview Inside each field, collapsed by default. Lists each value or range and the score it earns (value → score), ending with the score for no value, with a bar of the same bands.
The first concept opened. A PDF KEYWORDS block shows a keyword group name beside a wrap of grey keyword chips. An EXTRACTION FIELD 1 block shows the field name with a type badge that includes the unit, a sentence describing the extraction, and an open Score preview that lists each range with its score, ending with a no value line, above a segmented bar in shades of blue. A BLOOMBERG FIELD block below has a Key row with a field identifier. The second concept is closed.
Score preview starts closed; it is open here. The BLOOMBERG FIELD block appears only under the conditions described in When Bloomberg appears.

Adding or removing concepts

The pencil at the top right of the card opens Edit concepts.

ControlNotes
Concepts Opens a checklist of every concept you can use, each with its theme and a Search concept box that matches on name or theme. Click a row to tick or untick it. No concepts found appears when the search matches nothing.
Limit Up to 20 concepts. Ticking a twenty-first is refused with Up to 20 concepts can be selected.
N selectedRunning count of ticked concepts.
ApplySaves the selection and confirms with Saved — Concepts have been updated. The cost estimate updates.
CancelCloses the dialog.
The Edit concepts dialog with its Concepts list open. The field shows the chosen concepts' names separated by commas. Under it a panel has a Search concept box and rows with a checkbox, a concept name and its theme at the right. The panel covers the bottom of the dialog.
The checklist opens over the dialog's own buttons.

Editing a concept

A pencil on a concept's own row edits that concept in place. It does not appear on public concepts, which cannot be edited from here. It opens the same editor as the Concept Library, and saving it refreshes the agent.

Methods

How each company is checked against the concepts. Each method can be switched on or off. The card has no pencil icon: the time window every method shares is set on the General page, in Schedule.

The Methods page: a Methods card with five rows. Each has the method name, a round information icon and a switch at the right, all on and dark blue. The last row reads Data Feed - Bloomberg. The card has no pencil icon.
A method that is off is greyed out. Time window is no longer edited here; it is in the Schedule card under General.
MethodWhat it does
Keyword Count · Filings Counts how often your selected keywords appear across company filings — annual, interim and quarterly reports, earnings call transcripts, ESG reports, AGM materials, announcements and IR presentations.
Semantic Retrieval - Filings Uses Semantic Search to find filings that are contextually relevant to your keywords, even when the exact wording differs. Generally recommended for running against several thousand companies — references are provided but not to the depth of LLM Field Extraction.
LLM Field Extraction - Filings Uses Large Language Models to analyse and score Extraction Prompts across company filings, with context, sources and evidence statements to underpin each score.
LLM Field Extraction - News The same, across global news.
Data Feed - Bloomberg Pulls structured JSON data for your selected Bloomberg Fields from Bloomberg. Shown only when Bloomberg is set up for you — see When Bloomberg appears. Its switch saves like the others. It has no credit rate in the cost estimate, so when it is on it reads 0.00 credits in Cost breakdown.

Hover the info icon beside a method's name to read its description. The switch at the right of each row saves immediately; if it fails, the switch returns to where it was and Failed to update method appears. Each method except Data Feed - Bloomberg carries its own credit rate — see Estimated cost.

When Bloomberg appears

Bloomberg controls are shown only to a signed-in administrator who has saved a Bloomberg credential for the organisation. For everyone else — and whenever Orbit cannot read that state — they are hidden and the page behaves as before. That covers Data Feed - Bloomberg here and the BLOOMBERG FIELD section on Concepts. Open the page again after saving a credential to see them. Credentials are saved on Bloomberg Connection.

Time window

One window is shared by every method, and it is no longer edited on this page. Open General and use the pencil on the Schedule card: Edit Schedule has a Time window list next to Frequency.

Where the page leads

FromTo
View details (ESTIMATED COST)Credit
Share icon on a reportShare
Portfolio management link in Edit portfolio, and the save-result dialogPortfolio Management
Delete → Yes, DeleteAgent Marketplace
Get started (MCP/API Use)Orbit developer documentation, in a new tab