Ask
Ask is where you put a question to your own data and get the answer first: what the
run found, how many decisions wait for you and what they add up to. Open it with Ask,
the first entry under Fetch in the rail (/fetch). The app opens here after you
choose an instance.
From top to bottom, the page holds:
- The question box (Ask a question or use voice command…) and who the run is for.
- Start here, until your first run has finished.
- The plays and FETCH STARTERS: prepared questions, filtered by play.
- The answer for the selected run.
- How FACE knows this, folded: the sources, connections, coverage, runners and run history behind the answer.
Ask shows no money tiles. The money line (what is owed, at risk and waiting) is on Why (owner, 2026-11-02: the money tiles “are under Why page not Ask”).
Ask a question
Pick who the run is for
Beside the question box, Who this run is for shows Generalist view until you pick a persona, then For: {name}. The run is written for that persona: the plays lead with the persona’s plays, and the money line on Why leads with what the persona reads first. The persona changes what the answer leads with, never what FACE reads. Personas are set up under Settings › Personas.
Type, say or attach your question
Type in Ask a question or use voice command…, or pick a question under FETCH STARTERS. Beside the box:
- the microphone dictates your question;
- Camera Input takes a photo;
- Attach File attaches one or more files.
Attachments appear as chips you can remove. FACE reads each file by its content, whatever its name says:
- photos: their text, and for JPEG and PNG a structured reading of the document (type, fields, tables, checkboxes);
- PDF: the text of each page, and a scanned page as a photo;
- DOCX and PPTX: their text and tables; XLSX: its tables;
- plain text and HTML: the text;
- video clips: sampled frames, read the way a camera feed is.
A file FACE has no reader for is skipped, and the run says why. A run with only an attachment and no connection still answers, from what it read.
Choose how to be told
Call Me When Finished and WhatsApp Me tell you when the run ends: once every phase has ended, not when the answer comes back. The message says whether the run finished, partly finished, was interrupted by a server restart, or failed.
Run it
Press Enter or select the send button (Run Fetch). The activity card narrates the run, phase by phase, over whatever page you open next.
If the app cannot reach the server, the run is queued on your device and sent when the connection returns.
Start here
Until a run has finished, a Start here panel sits under the persona picker, above the plays: From no source to your first fetch. Each step is read from the platform, never ticked by hand. Its steps are Connect a data source, Have a runner ready and Run your first fetch. Each says whether it is done, or Could not check this step. The panel goes away once every step is done.
Plays and fetch starters
The plays are what FACE recovers or protects money on: All plays, Claims recovery, Disruptions, Routes, Fulfilment, Cold chain, Maintenance, Inventory and Fleet. Pick a chip to see only that play’s questions under FETCH STARTERS. The persona the run is for puts its plays first.
The starters come from your connections and your runs. While FACE samples your sources the row says Sampling your sources…, then Suggestions from a quick sample of your sources · {time}, how many sources answered, and Refresh. A source the sample could not read is named with a short reason; Show details shows the source’s own error.
If a play has no question yet, the row says so and offers Connect a source.
The answer
The answer card shows the selected run:
- its question, or when it ran, as the card’s title;
- who it was for: Generalist view or For: {name};
- the answer, laid out for the question (below);
- three ways on: Decide in Actionable Twins, See why and See the options.
With no run selected, the card is not shown: the question box and the plays are the page.
An answer laid out for its question
Each answer chooses its own layout. A question about late lanes can lead with the late shipments and the map; a “should I approve” question leads with the options; a “how much” question leads with the figure and what it was counted over. The first block always answers the question.
The model chooses which blocks to show and in what order. It never writes a figure: each block points at something the run measured, and FACE fills in the value from the run’s own records. The blocks are:
| Block | Shows |
|---|---|
| The answer | What the run said, checked against your data. |
| A figure | One measured figure, such as At risk, Delivered late or the rows breaking a rule, with the rows it was counted over. |
| A table | Rows from the run’s board, such as Late and past promise or Claims, with chosen columns and one filter: Late only, Closing within 14 days, Below safety stock, Missing evidence or Largest first. At most 8 rows. |
| A chart | A series from the run’s board, such as Shipments by status, as bars. |
| The map | The run’s routes, with Open the map. |
| Options | One decision’s what-if options, as See the options shows them. One is recommended only when it is a clear winner. |
| Decisions | What waits for a decision, or the cards the answer names. Each is decided in Actionable Twins, under its own approval rules. |
| Files | The files the run read, and how each was read. |
| What this run did not check | Phases that did not finish, sources not read and board sections not measured. |
Each block has How this was found: where its value was read, over which rows (a sample says so), and how it was computed.
A block whose source was read but holds nothing says Nothing to show: and why. A figure the rows do not state says Not measured. A block whose source could not be read says Could not be read: and the error. None of them is shown as zero.
How this answer was laid out says whether FACE chose the blocks for your question, or the standard layout stands in and why. It also lists any block the model asked for that the run does not hold, with the reason it was left out.
The standard layout is the answer, then how many decisions wait and what they add up to (No decision waiting when none does). It stands in when the model gave no layout, or one that named a value of its own instead of something the run measured.
The layout is stored with the run. Opening a past run shows the same layout, without asking the model again. A run from before layouts were stored shows the standard card: what it said (the spoken answer stays on the device that asked, so a run asked elsewhere reads The spoken answer stays on the device that asked. What the run left for you is below.), then what waits. If a run’s layout cannot be read, the card says so and shows the standard card.
The layout is ready when the run’s decisions are: until then, the card shows the standard answer.
How FACE knows this
How FACE knows this (Sources, coverage, runners and run history) is folded under the answer. Open it to see:
- how many of its sources the selected run read, naming the ones it could not read and why;
- how the run sampled each table and what the sample shows of each column, when the run recorded a sample;
- ACTIVE CONNECTIONS, then Operational coverage, then COMPUTE RUNNERS;
- EXECUTION HISTORY, with What the run also found and Runs recorded on the server.
Connections
Under ACTIVE CONNECTIONS, each connection has a card with its name and type, what the server holds for it (such as whether its credential is stored), and a caption: Included in fetch or Not included for an active connection, Switched off: not read by Fetch for one that is switched off. The caption describes your selection. It does not say whether the source is reachable.
Tap an active card to include it in, or leave it out of, your next run. New connection opens the connection wizard on Settings › Integrations › Systems. Connections a run proposed are listed with Confirm and Dismiss.
The Active switch
The Active switch on a connection or runner turns it on or off for every run on the instance: yours, other people’s and scheduled ones. The server enforces it and refuses a run that names a switched-off connection or runner.
Only an admin of the instance can flip the switch. Everyone else sees Active or Inactive. If a switch cannot be saved, {name} was not switched: {error} gives the reason. A switched-off connection also leaves the Map view’s live layers.
Live views
Cameras, sensors, field readers and devices are connection types. Their card opens a live view, already narrowed to that connection (Connection {id} only; Show all widens it):
| Connection type | Button | What it shows |
|---|---|---|
| Video Stream (cameras) | Surveillance | The Surveillance Dashboard: Feed Ingestion grades a feed (Threat Detected — {classification} or No Threat Detected), Sensor Cueing cues one sensor from another, and COP Telemetry Feed streams events. Photos taken with Camera Input are listed under Captured media. |
| MQTT Bus, Sensor Gateway, GPS Telematics, RFID Reader | Live sensors | What the always-on connections hear, the candidate findings, and for admins the health of the ingest path. |
| Runink Edge (a handheld field reader) | Field reader | That reader’s sightings as they arrive. When the stream stops, the last frame stays with its age. |
| Any source that reports devices | Devices | The devices the connection reported, with when each was last seen. |
A reading with no stated value is never drawn as a number, and silence nobody can judge is shown as unjudged, never as fine.
Operational coverage
Operational coverage shows what FACE can see of your operations in the sources it has read (fleet, yard, warehouse and racks, cold chain, shipments, orders, maintenance, drones and automation, IoT sensors and tags), and which decisions that supports. Each decision is Ready, Blocked or Couldn’t check.
Runners
Under COMPUTE RUNNERS, each runner has a card with its own Active switch. Tap an active runner to run your next question on it. The card marked Active is the runner the next run will use; other active runners show Standby, and a switched-off one shows Switched off. With no pick, the run goes to the managed runner if it is active, otherwise to the first active runner. New runner opens Settings › Integrations › Runners.
Execution history
What the run also found lists, for the selected run, the actions the judge removed from the answer (each marked with its verdict) and the forecasts, with ANOMALY where one was detected.
Runs recorded on the server lists every run the server recorded for you, newest first: when it started, how long it took, the runner, and how it ended. An org admin sees every run in the organization. The run cards above it are kept on this device.
| Outcome | Means |
|---|---|
| Successful | It answered, and every phase finished. |
| Partial | It answered, and at least one phase did not finish. The run names each one and why. |
| Running | It has not answered yet, or a phase after the answer is still running. |
| Answered | It answered, and the server has no record of its phases. It is never counted as successful. |
| Failed | It did not answer. |
| Refused | It was refused before it ran. |
| Interrupted | The server stopped before it answered. Run it again. |
The answer comes back before the run is done: the analysis (the checks, the rules and the scenario verdicts) and the twins’ decisions run after it, in the background.
Archive run history, for org admins, takes every run so far off every page (Ask, Briefs, Why, Options, Actionable Twins) for everyone on the instance. Nothing is deleted: the runs are kept for audit, and an admin can bring them back from Show archived with Restore or Restore all.
Open a past run
Tap a run under Runs recorded on the server to see its question and how it ended; Show this run’s results makes it the selected run.
Tap a run card kept on this device to open Fetch details (/fetch/run/<id>): how it
ended, Query, Runner, Connections, Started, Duration, Error when
there was one, Result, What happened to the sources, Automated analysis and
Generated SQL. Reuse request asks the same question again.
The page also shows What this fetch was given: each file or capture attached to the run and how it ended (read, read, nothing on it, skipped: {reason} or could not be read: {reason}). Show what was read opens the reading: the structured reading of a photo, Tables in the document, Text read (cut after 64 KiB, and it says so) and Left out. You can correct a value read from a document; the correction is recorded with your name and the value it replaced.
The selected run
Ask, Options, Why and Actionable Twins each show one run: the selected run. The page’s title line names it, with Change run and Clear under Run actions.
- A run you start selects itself when it answers.
- Otherwise the newest Successful run on the server is selected. A partial run is never selected for you, but you can open one.
- Open an earlier run lists every run the server recorded. Only a run with a transaction id can be opened.
- Clear selects no run. The choice is kept on this device.
When the selected run is not the newest one, a notice says so, for example This page shows the fetch of {date}; a newer fetch finished on {newer}., with Show the fetch of {date}.
With no run selected, Options, Why and Actionable Twins show only Your operation’s story starts with a fetch, with Run a fetch and Open an earlier run ({count}). A page shows what the selected run recorded and nothing else: no figures borrowed from other runs. If the run recorded nothing for a page, the page says so.
RPCs behind this page
FetchService.RunFetchStream(server stream),ListFetchStarters,ListFetchRuns,GetFetchAttachments,GetAskLayout(the answer’s stored layout).ConfigService.ListConnections,ListRunners,SetConnectionActive,SetRunnerActive(admin only).TwinsService.GetActionQueue, for the waiting decisions.HypothesisService.CompareOptionsandSendToTwins, when an options block is retried or an option is sent for approval.
See the API reference.