SORT AND ROUTE
Get requests to the right place.
Sort support messages, tag enquiries or choose a queue. Teach the difference between a new order, a delivery question and a return.
One message → one of your categories
Your choices.
Your examples.
Put to work.
Teach a local classifier how you sort, route and choose. Use it in the app, in your workflows, or as a decision tool for your agents.
Start with the Intent Library or your own labels. Review the results. Keep the work on your PC.
Available from Microsoft Store.
Part of the Unara family: built on the architecture we are developing for Unara. See what is coming
Decisions, with receipts.
- 01Define your choices
What should this workflow decide?
- 02Show it examples
Give each choice words from your world.
- 03Check, then connect
Use the result in a workflow you control.
LOCAL PROCESSINGNO GPU REQUIREDENGLISHNO USAGE METER
GIVE REPEATED DECISIONS A HOME
A useful answer
from choices you define.
Just Another Classifier helps turn incoming information into a choice your software can use. You set the labels, examples and next steps.
BUILD A WORKFLOW
Ask the next useful question.
Break a task into smaller decisions. Use known context to narrow the choices, then send uncertain cases to a person or a fallback.
A choice → a branch → a next step
WORK THROUGH A FILE
Make repeated work reviewable.
Process a batch, inspect the results and export them. Keep a local history so useful corrections can improve the next version.
Many records → results you can review
Working with numbers? Choose among candidates using the numeric attributes you supply. Meaning and Numbers questions can be part of the same Flow.
Want a second look? Set Reading for each Flow: Standard, Close Read or Second Opinion. Close Read and Second Opinion re-read each answer before it is given. When they are not convinced, the message is not answered, so nothing is done on a guess. Both are included and run on your PC.
START WITH ONE REAL TASK
Make it yours.
Then see how it does.
A Flow saves your questions, choices and workflow. Build it visually, with the help you need inside the app.
- 01 / CHOOSE
Find a starting point.
Search the Intent Library for a task or industry. Pick useful intents and example phrasings, edit them, and save your own Flow. You can also start from scratch.
- 02 / TEACH
Add examples that fit.
Each option needs at least 5 labelled examples; about 10 is a useful starting point. Include the details that separate similar choices. If your customers say what not to do, teach a few of those under the option they want. Add “None of these” examples for requests you want it to decline.
- 03 / PRACTISE
Find what needs attention.
Tune locally and review the choices it confuses. Check readiness on separate messages you have kept out of teaching. The report shows correct, wrong and declined results.
- 04 / IMPROVE
Learn from real use.
Review past answers, correct mistakes and save a new draft. Check it before making it live. Your examples and feedback stay under your control.
The example count is a starting point, not an accuracy guarantee. Clear choices and careful review matter. The full walkthrough is in the app under Help.
FOLLOW THE DECISION
An answer is the start.
The receipt tells the rest.
Recorded from the launch reader, 30 September 2026, taught with 5 examples per option.
RECORDED MESSAGE
Hi, the lamp I ordered last Tuesday still hasn't turned up and the tracking just says 'in transit'. Any idea when it will arrive? Thanks, Priya
ANSWER
Where is my order
Decision receipt
- Outcome
- answer
- Value
- Where is my order
- Top raw score
- 0.673
- Next option
- Change delivery address (0.319)
Scores are raw reader values, not probabilities.
RECORDED MESSAGE
Can I get a refund on my gym membership?
REFUSE
None of these
Decision receipt
- Outcome
- none
- Value
- null
- Top raw score
- 0.180
- Best option considered
- Return an item (0.180)
- Refused by
- threshold
Scores are raw reader values, not probabilities.
These are fixed recorded results, not a live request. The second message was refused because it did not fit the taught options.
Your work
stays local.
The app processes classification text on your PC. It does not send that text to us for a decision. You control local history, exports and connected applications.
A receipt
to inspect.
Keep the result and its reported evidence together. See the outcome and available decision details. A receipt helps you review an answer; it does not guarantee correctness.
A fallback
when needed.
Handle an answer, a refusal or a request for clarification explicitly. An unsure result can go to review instead of being treated as a confident instruction.
A draft
before live.
Use the visual canvas to edit questions, conditions and fallbacks. Preview changes, inspect results and choose when a saved version becomes live.
FOR YOUR APPS AND AGENTS
A local decision tool
in a bigger workflow.
Connect through the local API, command line, SDK or a compatible MCP client. Use the app's Help for setup instructions that match your version.
Let an agent use it.
An agent can send a message to a saved Flow and receive a structured result. Use that result to choose a queue, branch or review step within the permissions of your application.
Let an agent help teach it.
Through the teaching API, an agent can help organise examples, review feedback and check a draft. Give it only the access it needs, and keep approval to go live separate.
Keep actions under your control.
A classification is data, not permission to send a message, change a file or spend money. Your application decides which tools, destinations and actions are allowed.
A local classifier does not make a connected cloud agent local. Check that agent's data handling before sharing your material with it.
Agent setup overviewBEFORE YOU INSTALL
Built for your PC.
Windows first. A macOS version is planned.
- Windows
- Windows 10 22H2 or later, or Windows 11, on an x64 PC.
- Processor
- Any 64-bit (x64) processor. Classification runs on your CPU, no graphics card needed. Four or more cores recommended; CPU use can be adjusted in the app.
- Language
- English. Support for other languages is planned for later.
- Connection
- Internet access for download, checkout, activation, licence renewal and updates. Classification runs locally.
- Offline use
- A valid cached licence supports up to 3 days offline, capped by the access date shown in the app. Help remains available offline.
- Memory and storage
- At least 4 GB of RAM, 8 GB recommended. The classifier engine uses under 1 GB while it runs with Reading set to Standard; Close Read adds up to about 0.5 GB and Second Opinion up to about 1 GB while in use. Plan for at least 3 GB of free disk space before installing: the download is about 1.1 GB and the installed app about 1.5 GB. Installation, updates and your saved data may need more.
SIMPLE, PER-DEVICE PRICING
One subscription.
No per-request charge.
MONTHLY
US$2/month
One device. Your local decisions.
YEARLY
US$20/year
One device. A year of local decisions.
Per device. 3-day free trial with a card up front. Your subscription continues at the selected price unless you cancel. Cancel in two clicks in the app. Taxes may apply.
Buying for a team? Contact us.
A few clear answers.
What should I try first?
Choose one repeated task you already understand, such as routing a support message. Keep the options distinct, start from the Library or your own examples, then try separate messages and inspect the results.
Do I need to be a developer?
You can create and review Flows in the visual app. Developers and compatible agents can connect to the same local service. Detailed setup instructions stay in the bundled Help.
How many examples do I need?
At least 5 per option, with about 10 recommended to start. More examples help when they add useful variety or show the difference between close choices. Check on separate messages before trusting a workflow.
What if a message says not to do something?
When it recognises that a message says not to do the selected action, for example “please don’t cancel it”, the Flow refuses instead of answering with that action. Wording varies, so teach a few examples of how your customers say it under the option they want. For extra care, set Reading to Close Read or Second Opinion.
What are Close Read and Second Opinion?
Two Reading options that re-read each answer before it is given. Close Read is quick. Second Opinion uses our largest reader and is strongest on the hardest cases. When either is not convinced, the message is not answered. They are readers we tuned for this job from Microsoft’s open DeBERTa-v3 models (MIT licence), and they run on your PC.
What happens when it is unsure?
Inspect the result and receipt, ask for more context or send the case to review. Refusal and clarification are outcomes your workflow can handle. Errors such as a lost connection are kept separate.
Does it execute the actions in my Flow?
A Flow returns action data. Your connected application decides whether to perform an action and enforces its own permissions. Keep consequential actions behind appropriate review.
Can I work without an internet connection?
Classification runs locally. The app connects for account services and updates. A valid cached licence supports up to 3 days offline, capped by the access date in the app. Your installed Help remains available offline.
Where does my text go?
The app does not upload classification text to us. History stays on your PC. You choose what to share through exports or connected tools. Account, activation and billing use separate online services.
Why is Sign in greyed out?
Select Refresh sign-in status on the Account page, then Sign in. This can happen just after the app opens, especially after a PC has been released and signed in again several times. One refresh is usually enough.
How do I cancel?
Open Account in the app, choose Cancel subscription, then confirm. Check the resulting access end date. You can also manage your subscription in the account portal.
BRING ONE TASK
Give it your choices.
See what you can build.
Start with the basics, then use the version-matched guide inside the app.
We keep improving Just Another Classifier, and every release is measured and published. Thank you to every customer who helps us make it better.