YOUR SERVERYOUR MODEL
Your server
We send the photo to your server and read the answer back: what it is, and how sure.
- RUNS ON
- your server
- MODEL
- the one you already run
- SIGNAL
- needed; photos wait until online
- IT TELLS YOU
- what it is, and how sure
AI & COMPUTER VISION
Put a vision model behind one form question.
Add a vision question to a form. When a worker takes a photo, it is classified by your model, a managed one, or on the phone, and the result lands on the record.
Ask it what a photo shows, where things are in it, or both. The answer is saved on the record along with which model gave it.
GOODS RECEIPT · Q4 · PHOTO OF THE GOODS
FlipPHOTO OF THE GOODS · Q4
WHAT THE MODEL FOUNDAI Classification
completed- Label
- Tape roll
- Confidence
- 94%
- Source
- your model
- Model
- goods-detect v3
OBJECTS
- Tape roll · 94%
- Carton · 78%
ONE PHOTO QUESTION · THE ANSWER LANDS ON THE RECORD
BRING YOUR OWN MODELAlready have a vision model?
Already have a vision model?
Point a question at it.
Every vision question runs in one of four modes. Two are for teams that already have a model, and it stays yours. If you don't have one yet, we'll help you build one for your products.
HOSTED BY USYOUR MODEL
Your model, hosted by us
Your model runs on TraceSource servers and handles each photo. You bring it and we do the serving, or our team trains it for you.
- RUNS ON
- Us
- MODEL
- the one you send us
- SIGNAL
- needed; photos wait until online
- IT TELLS YOU
- what it is, where it is, or both
ON THE PHONE
On the phone
A small model on the phone labels what the camera sees, so the result needs no signal at all.
- RUNS ON
- the phone
- MODEL
- a small one kept on the phone
- SIGNAL
- not needed
- IT TELLS YOU
- what it is, where it is, or both
MANAGED FOR YOU
Managed
We classify the photo for you with a managed model. There's nothing to bring or host.
- RUNS ON
- our servers
- MODEL
- our model
- SIGNAL
- needed; photos wait until online
- IT TELLS YOU
- what it is, where it is, or both
IN THE FORM
It's just another question.
Add a vision question next to the ordinary ones. The worker takes a photo the way they always do, and if they're offline it waits with the rest of the form. When it runs, the answer says what it saw and how sure it is.
Goods receipt
FORM · 4 QUESTIONS
- TEXTSite
- NUMBERQuantity received
- SCANPallet label
- PHOTO · AIPhoto of the goods
AI CLASSIFICATION
Task
Object detection
Instructions
Count the tape rolls; flag torn wrap.
Keep the full result
YOUR MODEL
Your serverOn the phone
Server address
https://api.example.com/infer
Access key
Key in header
Header name
X-API-Key
Key •••••••• · saved
ONE PHOTO, STEP BY STEP
- 09:12:41Photo captured · queued offlinepending
- 09:31:07Back online — sent to your serverrunning
- 09:31:09Label stored on the answercompleted
How sure the model must be
You set the bar per question. Anything below it is flagged for a person to check.
YOUR BAR · 60%48% · flagged94% · kept
ON THE PHONE
How your model gets onto the phone.
Upload your model once. Each phone fetches it over a secure link, checks it arrived intact, and remembers which version it has, so every answer says which model produced it.
HOW YOUR MODEL REACHES THE PHONE
Upload your model
v1v2 ✓ activeone file · 8.4 MB · version 2
The phone asks for it
the phone fetches it when the form opens
A secure link that expires
the download is checked before it is used
It runs inside the camera
labels what it sees, with no signal needed
EVERY ANSWER RECORDS WHICH MODEL VERSION PRODUCED IT
No connectionHome good(84%)unclassifiedHome good (84%) · 1 unclassifiedCAMERA · IMAGE · VOICEOn the phone
- NEEDS NETWORK
- No — it runs on the phone
- LABELS
- says what it sees, or unclassified
- MODE
- live in the camera
Photo of the goods
Queued offlineSent to your server on sync. Label and confidence land on this answer.
SubmitOn a server
- NEEDS NETWORK
- Yes — the photo queues offline
- RUN
- queued → running → done
- MODEL
- yours, or the managed one
DESCRIBE IT. GET A FORM.
Describe the form and the assistant drafts it.
Say what you need in plain words and the assistant drafts the questions. Ask for another field or change one in chat, and the form is ready to fill in the web builder and in the mobile app.
- 01Drafts questions from a description
- 02Adds sections and questions on request
- 03Same assistant in the mobile app
YOU · IN CHAT
"Draft a site inspection form: site, date, a photo of the meter and the reading."
Site inspection
DRAFTED BY THE ASSISTANT · 5 QUESTIONS- TEXTSite namerequired
- DATEInspection datedefaults to today
- CAMERAPhoto of the meterAI vision · classification
- NUMBERMeter readingadded on request
- TEXTAREAIssues foundoptional
ASSISTANTdrafted in one go
Drafted “Site inspection” with 5 questions and added “Meter reading” after the photo, as asked.
USAGE & CREDITS
Every plan comes with credits for AI usage.
AI usage is metered in credits and comes out of your plan's allowance. Form drafting and automations count towards it.
AI CREDITS · THIS MONTH
412/ 1,000
MAX PER RUN 120RECENT RUNS
- DRAFTGoods receipt · form drafted in chat26 SEPcompleted
- AUTOMATIONLow stock alert · ran on state change25 SEPsucceeded
- DRAFTSite inspection · 5 questions24 SEPcompleted
- AUTOMATIONOverdue work orders · daily23 SEPfailed
STARTER · MONTHLY1,000120 per run
PRO · MONTHLY5,000250 per run
ENTERPRISECustomset per contract
WHERE IT RUNS · VISION IS THE PHONE CAMERA; THE REST SCAN OR PRINT
Phone camera
VISION · SCANNING
CSL CS108
UHF RFID · SCANNING
Specifications ↗Urovo R70
WEARABLE · SCANNING
Specifications ↗SATO WS4
LABELS · PRINTING
Specifications ↗
WHAT IT ISN'T
What it does not do.
Four limits, said up front so nobody has to find out later.
The platform does not train models.
Bring your own, use the managed option, or ask us. We help teams build vision models for their products.
There is no model marketplace.
The model is yours, ours, or a small one on the phone.
We don't publish accuracy figures.
Those come from your model. You set how sure it must be, and review what falls below.
A classification is a suggestion on the record.
People still sign off.
READY WHEN YOU ARE
Put a camera on your form.
Add the question, choose the mode, and the label lands on the record.
We'll wire up your model, or help you build one for your products.