Use your vocabulary
Choose labels such as Billing, Account access and Delivery. Add short definitions to make their boundaries clear.
Zero-shot text classification
Text in.
Decisions ready.
Turn messages, reviews and documents into categories your software can use. Define the labels. Send your text. Put the result to work.
Your categories. No labelled training set required.
SUPPORT ROUTINGVerified example
01 / Your text
02 / Your candidate labels
BillingAccount accessDelivery
03 / Selected category
Billing128 classesMaximum candidates per task
1,024 textsPer request, within size limits
2 modesSelected label or label with scores
A small decision. A useful next step.
Give your workflow a clear answer.
Route a customer message, organize feedback, or sort a collection of documents using the categories that matter to you.
Ask one clear question
Send one text or a batch with the same labels and instruction. Each text receives its own result.
Make the next move
Use the selected label to choose a queue, group a report, or take the next step in your existing code.
From text to action
See what a category can do.
These results were checked against the service. They illustrate the workflow; they are not guarantees for every input.
Route customer messages
Turn incoming requests into categories your workflow can route.
BillingAccount accessDelivery
“I was charged twice for the same order.”Billing
“Please help me reset my password.”Account access
“Where is my parcel? It has not arrived.”Delivery
Example mode: choice. Checked 2026-09-18.
Organize customer feedback
Group reviews by sentiment so you can explore patterns in feedback.
PositiveNegativeNeutral
“The replacement arrived quickly and works perfectly.”Positive
“It broke on the first day and support never replied.”Negative
“The package contains a cable and an instruction booklet.”Neutral
Example mode: softmax. Checked 2026-09-18.
The result you need
One decision. Two ways to read it.
Pick a category
A selected label from your list. A simple result for routing, grouping and branching.
mode: "choice"“champion” is an alias for this mode.
Compare the candidates
A selected label plus relative scores for every category, so you can inspect how the choices compare.
mode: "softmax"Scores are uncalibrated. Evaluate them on your own data before choosing thresholds.
From one message to a worksheet
Give it the whole batch.
Classify up to 1024 texts per request, within the request-size limit. Group rows that share the same classification task and labels.
Follow the Python workflow →- Results stay in input order
- A status for every individual result
- Python helper with no extra packages
- Batch sizes handled by the client
Your next step
Start here. Build it your way.
Integration guide
Authentication, batching, errors, and a complete CSV-to-results Python workflow.
Read the guide →API reference
Inspect the request and response fields, then run a classification in your browser.
Open the reference →Ready for your clanker
The same guide as Markdown, the API schema, and a downloadable Python client.
Get the agent resources →The integration guide and API reference ask for your key once per browser tab. Paste just the key value; the pages handle authentication and downloads for you.
Clear categories.
Clear expectations.
Choose between 2 and 128 labels. Every successful result selects one of them. Include “Other” if you need that outcome. Classification can be mistaken, and candidate order can affect predictions; test representative examples before relying on decisions.
Inspired by TypeSafe’s Jev, FlexiClass targets the same classification and routing outcomes. It has its own API; this comparison does not promise identical predictions or API compatibility.