01
Upload your examples.
A labeled TXT file is all you need. Add one example and its 0 or 1 label on each line.
1 | I approve this action.
0 | Do not proceed.
1 | I accept.
THE DECISION API FOR SMALL TEAMS
One decision.
Zero excess.
Route forms and messages to the right workflow.
Train on your examples. Measure on separate data.
Return just 0 or 1 to your app.
TXT training and testing · Current pricing in your console · Free output
NULIQ / DECISION CORE01—00
ONE BIT. THAT’S IT.
INPUT → CLASSIFY → DECIDE0 / 1
SAMPLE INPUT
“I approve this action.”
1
YESIllustrative example
0 / 1One clear decision
.TXTYour training data
APIReady for your app
$0Output cost
01 / YOUR MODEL. YOUR RULES.
Keep your answers short.
Make your decisions clear.
Make text classification a building block. Let your examples define what “yes” and “no” mean.
02
Train. Test. Improve.
Train from your console. Measure errors with a separate TXT file. Activate the new version when you are ready.
v1v2v3
03
Connect through one API.
Send text from your server and get one bit back. API access, Playground, and model management share one account.
POST /v1/decide1text/plain
Built for decisions. NULIQ is a binary text classifier. The starter model was trained on Turkish approval and rejection examples; it is not a general knowledge model. Train and validate on separate examples for your own task and language.
CHECK QUALITY BEFORE YOU CONNECT
See the wrong decisions.
Then connect your workflow.
For small teams managing forms and customer workflows: uploads, evaluation, version selection, and an API in one console.
YOU DEFINE THE DECISIONRoute 1
Route 1
Other workflow 0
For example, label a message to be passed to sales as 1 and other messages as 0. Correct labels and representative test data depend on your process.
01
Train on your examples.Create a separate version for each customer and decision task.
02
Test on separate data.See false positives and false negatives separately. Tests that overlap with training data are rejected.
03
Measure, then activate.Your API address stays the same. If a new version does not work for you, select a previous one.
A pretrained model is not an accuracy guarantee. Alongside the Turkish approval/rejection starter model, Guard EN is a narrow spam example trained on historical English SMS data. It does not establish performance on Turkish messages or web forms. Test independently on your own data.
02 / LESS CODE. CLEAR OUTPUT.
Your decision layer.
In every app.
Connect from your server with the REST API. A compatibility endpoint is also available for supported clients using the Ollama API format.
01
Send JSON.
Get 0 or 1 as plain text.
02
Update your model.
Keep your integration.
03
Measure the input.
Pay nothing for output.
REST API / cURL
# One input. One decision.
curl $NULIQ_URL/v1/decide \
-H "Authorization: Bearer $NULIQ_KEY" \
-H "Content-Type: application/json" \
-d '{"input":"I approve this action."}'RESPONSEtext/plain
1Example output · The response body is a single character.
The Ollama connection supports an API subset. It is not a local Ollama model file and does not support tools, images, or chat history.
03 / PAY FOR WHAT YOU SEND
Small decisions.
Transparent pricing.
Loading current pricing from the server. Input is measured in UTF-8 bytes; output is free.
INPUT PRICINGLoading pricing
—/ 1 MB of input
1 MB = 1,000,000 bytes · USD
- Output cost $0
- TXT training and independent testing Included
- REST + Ollama API subset Included
- Maximum query size 4,096 bytes
Every bit accounted for.
Example usage
1,000queries × 100 bytes
100,000 input bytes—
Some characters use more than one byte. The console shows the exact input size and estimated cost before you send a query.
API and Playground use the same pricing. Failed or invalid queries are not charged. Training and quality tests do not debit your query balance; usage limits apply.
Before paying, review the terms of use, refund and cancellation policy and service delivery information.
Try the pretrained starter model.
Improve it with your own data as your needs become clear.