Sign in
Open OAE Studio and sign in or create an account. Your account keeps server OAEs connected across sessions.
OAE Studio Wiki Production guide
Teach useful AI.
Keep it understandable.
OAE Studio helps you create focused AI entities—OAEs—that learn from examples you provide. Training and testing happen securely through OAE Studio services, while the Windows app keeps your workflow clear and approachable.
Create
Select +, name the OAE, choose its type, and add a short description so its purpose stays obvious.
Teach
Add representative examples. Quality and variety matter more than repeating nearly identical phrases.
Train & test
Train while online, then try realistic inputs that were not copied directly from your examples.
✓
A good first result
Your OAE responds correctly to new wording—not only the exact examples it saw during training.
Choose the right tool
Three OAE types, three kinds of answers
Start with the output you need. You can create several small, focused OAEs instead of forcing unrelated behavior into one model.
Text / Chat
Matches a user message to the most relevant taught response.
- FAQ assistants
- Guided help
- Character dialogue
Classifier
Assigns text to one of the categories you define.
- Request routing
- Topic detection
- Intent labeling
Decision AI
Chooses an action from structured facts and shows the decision path.
- Game behavior
- Workflow choices
- Rule-like guidance
| Type | Input | Output | Best when |
|---|---|---|---|
| Text / Chat | Natural language | Taught response | Several phrasings should lead to a useful reply |
| Classifier | Natural language | Category + scores | You need a label, route, or intent |
| Decision AI | Named structured values | Action + path | A choice depends on numbers, booleans, or categories |
Training guide
Text / Chat
Teach pairs of likely inputs and the response the OAE should return. Include natural variations that express the same intent.
Example datasetAtlas
hello→Hey! How are you?hi there→Hey! How are you?who are you→My name is Atlas.Make examples distinct
Use different wording, word order, and common abbreviations. Avoid adding dozens of tiny punctuation variations.
Use a confidence threshold
The threshold controls how certain the OAE must be before returning a match. Raise it to reject uncertain requests; lower it carefully when natural variations are missed.
!
Test unseen wording
After training, try phrases such as “who r u” rather than testing only exact training inputs.
Training guide
Classifier
Give each example a category. A useful classifier needs multiple representative examples per category and clear distinctions between categories.
Example categoriesRequest router
Supportcannot log intechnical errorreset access
Salesshow available plansrequest a demopricing question
Billingincorrect chargeinvoice questionpayment failed
- 1Define categories first
Use labels that describe one clear outcome and do not overlap unnecessarily.
- 2Balance the examples
Give every category enough varied examples so one does not dominate by volume alone.
- 3Review close scores
When top category scores are close, treat the result as ambiguous and add better examples.
Training guide
Decision AI
Decision AI learns from structured examples. First define a feature schema, then provide complete feature values and the action that should result.
NUMBERHealth
+12BOOLEANEnemy visible
→trueACTIONRetreat
confidence + pathFeature types
- Number for measurable values
- Boolean for yes/no facts
- Category for one value from a defined set
Read the explanation
Results include the predicted action, supporting sample count, action distribution, and path through the trained tree. Use these details to spot thin or surprising branches.
i
Schema changes affect training data
When changing a feature, review existing examples for compatibility and retrain before testing.
Your workspace
Account & sync behavior
Your signed-in account determines which OAEs you can access. OAE metadata, training data, and trained models are associated with that account on OAE Studio services.
Secure sessions
The app keeps a protected refresh credential on your Windows device. Signing out removes the local session.
Server source of truth
After reconnecting or signing in again, the app reloads your current OAE list and metadata from the service.
Truthful offline state
Cached metadata can help you recognize your workspace while offline, but editing, training, and inference wait for a connection.
Connectivity
Why training requires the server
OAE Studio’s Windows app is the workspace and editor; authoritative AI execution happens through OAE Studio services. This keeps model versions consistent between sessions and avoids quietly using a different local model.
Windows appEdit examples · inspect results
secure connectionOAE Studio servicesTrain · version · test
- The app clearly reports when a connection is unavailable.
- Training and testing never silently fall back to an old local engine.
- A successful training run creates a new server model version.
- Changing training data marks the current model as needing retraining.
Trust
Privacy & data behavior
OAE Studio processes the information you intentionally add to your account and OAEs. Treat training examples like any other cloud-synchronized application data.
What is synchronized
OAE names, descriptions, schemas, training examples, model metadata, and account session information.
What stays protected
Account credentials and session credentials are not shown in the interface or included in ordinary exports.
What not to add
Avoid passwords, authentication codes, private keys, payment-card details, or unnecessary sensitive personal information.
Legacy projects
Legacy import uploads validated training data for server retraining. Old local model files are not uploaded as authoritative models.
Support
Troubleshooting
OAE Studio says it is offline
Confirm the device has internet access, then use Settings to check the service connection. Corporate proxies, VPNs, or filtering software may need to allow OAE Studio’s public services.
Training says more examples are required
Add the requested minimum and ensure the dataset contains the necessary variety. Classifier and Decision AI projects require at least two distinct outcomes.
An OAE says “Needs training”
The dataset, schema, or a model-affecting setting changed after the current model was produced. Train again to create a fresh version.
A result is uncertain or incorrect
Add examples that clearly demonstrate the intended distinction. Remove duplicates and contradictory examples, retrain, then test with unseen input.
Sign-in no longer works on this device
Sign in again. A session may have expired or been revoked. If the problem continues, verify the service connection and your system clock.
Support
Frequently asked questions
What does OAE mean?
OAE is the product’s name for a focused AI entity you create and teach for a particular job.
Can I train while offline?
No. Training and inference require a connection to OAE Studio services. The app does not silently substitute local AI.
Do I need to write code?
No. Training examples and Decision AI features are managed through the Windows interface.
Why are there model versions?
Each successful training run produces a version so the model metadata and training history remain understandable.
Can another account see my OAEs?
No. OAE access is scoped to the authenticated owner.
Does legacy import reuse my old model?
No. It imports compatible training material and asks the server to create a new authoritative model.
✦Back to start ↑
Build small. Teach clearly. Test honestly.
The most reliable OAE starts with one well-defined job and examples that represent real use.
No documentation found
Try a broader term such as “training,” “offline,” or “Decision AI.”