A Windows-native offline workspace for turning customer conversations into reviewed intent datasets, local ML.NET intent classifiers, evaluations, recommendations, and business reports.
IntentOps Studio is a local Windows desktop application for importing chatbot conversations, reviewing intent samples, building datasets, training local classifiers, evaluating model quality, and generating actionable recommendations without depending on cloud services.
IntentOps Studio helps teams understand what users are asking, where chatbot coverage is weak, and what needs to improve before an assistant, chatbot, support flow, or intent classifier can be trusted in production.
Everything runs locally. Imported conversations, reviewed samples, datasets, trained models, evaluations, recommendations, and reports stay on the user's computer.
- Windows-native desktop app built with WPF and .NET 8.
- Offline-first workflow for intent operations and chatbot quality analysis.
- Local project workspaces with optional encrypted project databases.
- Business Profile editor for business context, products, glossary, intents, and routing rules.
- Conversation import flow using structured JSON exports.
- Customer-message sample extraction from imported conversations.
- Human review workflow for confirming, correcting, marking unknown, or discarding samples.
- Dataset builder with minimum examples, maximum samples per intent, weak-intent filtering, and class balancing.
- Dataset export as JSON, JSONL, CSV, and IntentOps dataset package.
- Local ML.NET text classification training.
- Training run history and model family workflow.
- Model export/import using IntentOps model package files.
- Duplicate dataset and duplicate model import protection.
- Evaluation screen with accuracy, Macro F1, Micro F1, LogLoss, confusion matrix, and per-intent metrics.
- Recommendations engine for surfacing weak intents, coverage gaps, and improvement actions.
- Recommendations and gaps export as JSON.
- Reports screen for Import Analysis, Dataset Summary, Model Evaluation, and Recommendations reports.
- Recent reports browser.
- Project backup and restore support.
- SQLCipher-oriented encrypted database workflow for local project data.
- Dark polished WPF interface using MaterialDesignThemes.
- Self-contained Windows release build support.
- Installer-friendly output for tools such as InstallBunker.
IntentOps Studio is useful when you need to answer questions like:
- What are customers actually asking?
- Which intents are underrepresented?
- Which intents are confusing the model?
- Which chatbot flows need more examples?
- Which samples should be reviewed before training?
- Which model version performed better?
- Which reports can explain the current state of the assistant?
- What should be improved next?
It is especially useful for:
- chatbot dataset preparation;
- intent classification experiments;
- offline AI workflow demos;
- customer support conversation analysis;
- business-domain intent mapping;
- internal assistant evaluation;
- model-quality documentation;
- portfolio and proof-of-concept AI tooling.
IntentOps Studio includes these main screens:
| Screen | Purpose |
|---|---|
| Splash Screen | Starts the application and shows the product identity. |
| Welcome | Creates, opens, restores, and manages local projects. |
| New Project | Creates encrypted or unencrypted local workspaces. |
| Project Dashboard | Shows project progress and step-by-step workflow state. |
| Business Profile | Defines business context, products, glossary, intents, and routing rules. |
| Import Conversations | Imports structured conversation exports into the project. |
| Review Samples | Lets humans confirm, correct, mark unknown, or discard extracted samples. |
| Dataset | Builds, imports, exports, and packages reviewed intent datasets. |
| Training | Trains local intent classifiers and manages model families/runs. |
| Evaluation | Displays model metrics, confusion matrix, and per-intent performance. |
| Recommendations | Shows improvement actions and intent gaps. |
| Reports | Generates business-readable HTML reports. |
| Settings | Stores local app preferences and project-related settings. |
A project is a local workspace that stores all imported conversations, business profile data, reviewed samples, datasets, trained models, evaluations, recommendations, and reports.
Projects can be created with local database encryption enabled.
New Project
├─ Name
├─ Description
├─ Language
├─ Industry
└─ Optional encrypted local project database
The Business Profile gives IntentOps Studio context about the company, product, domain, vocabulary, expected intents, and routing priorities.
Business Profile
├─ Business information
├─ Products / services
├─ Glossary
├─ Intents
└─ Routing rules
This makes intent review and recommendations more meaningful.
IntentOps Studio imports structured conversation JSON files.
A conversation export contains:
Source metadata
Business information
Conversation list
Customer messages
Agent messages
Outcomes
Channels
Timestamps
The app extracts useful customer messages and prepares them for review.
The review step is where raw customer messages become reliable training data.
Each sample can be:
| Action | Meaning |
|---|---|
| Confirm | Accept the suggested intent. |
| Correct | Change the intent manually. |
| Unknown | Keep the sample as unresolved or unclear. |
| Discard | Remove noisy or low-value samples from dataset creation. |
This human-in-the-loop step is one of the most important parts of the workflow.
Reviewed samples are turned into a dataset.
The dataset builder supports:
- minimum examples per intent;
- max samples per intent;
- weak-intent filtering;
- class balancing;
- dataset name;
- dataset notes;
- dataset history;
- JSON export;
- JSONL export;
- CSV export;
- dataset package export/import.
Dataset packages are protected against duplicate imports.
IntentOps Studio trains a local intent classifier from the selected dataset.
Training supports:
- brand-new model creation;
- adding a new training run to an existing model family;
- model naming;
- test split configuration;
- deterministic seed;
- max iterations;
- minimum sample rules;
- training notes;
- model history;
- model package export/import.
Model packages are protected against duplicate imports.
Evaluation shows the quality of the selected model.
Metrics include:
| Metric | Meaning |
|---|---|
| Accuracy | Overall prediction correctness. |
| Macro F1 | Balanced per-intent F1 average. |
| Micro F1 | Global F1 across samples. |
| LogLoss | Prediction confidence/loss signal. |
| Confusion Matrix | Which intents are being confused. |
| Per-intent Metrics | Quality indicators for each intent. |
Small datasets may produce unstable metrics. IntentOps Studio makes that visible so the dataset can be improved.
Recommendations convert evaluation and dataset signals into practical improvement actions.
IntentOps Studio can surface:
- weak intents;
- low-support intents;
- confusing intents;
- dataset imbalance;
- model-quality warnings;
- next-step suggestions;
- gaps export;
- recommendations export.
A recommendation such as:
Improve intent: Payment Issue
means that the intent likely needs more reviewed examples, clearer samples, or better coverage before the model can reliably classify it.
Reports produce readable HTML outputs for project documentation.
Current report types include:
| Report | Purpose |
|---|---|
| Import Analysis | Summarizes imported conversations and message structure. |
| Dataset Summary | Summarizes dataset composition and intent distribution. |
| Model Evaluation | Documents model metrics and evaluation status. |
| Recommendations | Summarizes improvement actions and detected gaps. |
Reports are stored locally and can be opened from the recent reports list.
IntentOps Studio uses structured local files for interchange and backup.
Business Profile JSON describes:
business
products
glossary
intents
routingRules
Conversation export JSON describes:
format
version
source
business
conversations
messages
outcome
Datasets can be exported as:
| Format | Purpose |
|---|---|
| JSON | Full structured dataset metadata and samples. |
| JSONL | Training-friendly line-based format. |
| CSV | Spreadsheet-friendly dataset view. |
| IntentOps Dataset Package | Portable dataset archive for import/export. |
Models can be exported as:
| File | Purpose |
|---|---|
| IntentOps Model Package | Portable model archive. |
| Model manifest | Metadata, metrics, dataset id, intents, and model type. |
| Model binary | Local trained model artifact. |
IntentOps Studio stores project data locally on the user's machine.
Typical stored data includes:
Project metadata
Business profile JSON
Imported conversation database
Reviewed samples
Datasets
Models
Evaluations
Recommendations
Reports
Backups
Encrypted projects require the correct project password to open their local database.
IntentOps Studio cannot recover lost project passwords.
IntentOps Studio is designed for local workflows.
- Conversation data stays on the user's machine.
- Training runs locally.
- Evaluation runs locally.
- Recommendations are generated locally.
- Reports are generated locally.
- Project databases can be encrypted.
- No external AI API is required for the main workflow.
Users are responsible for handling imported customer data according to applicable privacy, security, and compliance requirements.
WPF application layer
│
├─ Views
├─ ViewModels
├─ Dialogs
├─ Navigation
├─ UI services
└─ App composition
│
▼
Contracts
├─ Models
├─ Schemas
├─ Options
└─ Shared abstractions
│
▼
Core
├─ Project persistence
├─ Business profile handling
├─ Conversation import
├─ Sample extraction
├─ Dataset building
├─ Local model training
├─ Evaluation
├─ Recommendations
├─ Reports
├─ Packaging
└─ Backup / restore
The project is split into three main assemblies:
| Project | Responsibility |
|---|---|
IntentOps.App |
WPF UI, navigation, dialogs, view models, themes, startup, and desktop services. |
IntentOps.Contracts |
Shared models, schemas, interfaces, and contracts. |
IntentOps.Core |
Import, persistence, dataset, training, evaluation, recommendations, reports, and packaging logic. |
- C# / .NET 8
- Windows Presentation Foundation (WPF)
- ML.NET
- SQLite / SQLCipher-oriented local persistence
- CommunityToolkit.Mvvm
- MaterialDesignThemes
- MaterialDesignColors
- Microsoft.Extensions.Hosting
- Microsoft.Extensions.DependencyInjection
- Microsoft.Extensions.Logging
- Microsoft.Xaml.Behaviors.Wpf
For the installed release:
- Windows 10 or Windows 11
- x64 operating system
- Local disk access for project workspaces
For development:
- Windows 10 or Windows 11 x64
- Visual Studio 2022 or newer
- .NET 8 SDK
- .NET desktop development workload
Steps:
- Clone the repository.
- Open the solution file in Visual Studio.
- Restore NuGet packages.
- Build the solution.
- Run the WPF application with
F5.
Command line:
dotnet restore .\IntentOps.slnx
dotnet build .\IntentOps.slnx --configuration Debug
dotnet run --project .\IntentOps.App\IntentOps.App.csprojRun the automated test suite with:
dotnet test .\IntentOps.slnx --configuration DebugRecommended pre-release validation:
dotnet restore .\IntentOps.slnx
dotnet build .\IntentOps.slnx -c Release
dotnet test .\IntentOps.Core.Tests\IntentOps.Core.Tests.csproj -c ReleaseSuggested manual smoke test:
- Create an encrypted project.
- Close and reopen the app.
- Unlock the project.
- Import a Business Profile.
- Import conversations.
- Review samples.
- Build a dataset.
- Export dataset JSON, JSONL, CSV, and package.
- Train a local model.
- Export and re-import the model package.
- Run Evaluation.
- Generate Recommendations.
- Export recommendations and gaps JSON.
- Generate all reports.
- Backup and restore the project.
Recommended Windows release publish command:
dotnet publish .\IntentOps.App\IntentOps.App.csproj `
-c Release `
-r win-x64 `
--self-contained true `
-p:PublishSingleFile=false `
-p:PublishReadyToRun=true `
-p:PublishTrimmed=false `
-o .\publish\IntentOps-win-x64Recommended release settings:
| Option | Value |
|---|---|
| Target framework | net8.0-windows |
| Runtime | win-x64 |
| Self-contained | Yes |
| App single file | No |
| ReadyToRun | Yes |
| Trimming | No |
| Native AOT | No |
| Installer single file | Yes, if using an installer packager such as InstallBunker |
The app publish output should remain multi-file for maximum compatibility with WPF, MaterialDesignThemes, ML.NET, SQLite/SQLCipher-related dependencies, and native assets.
The installer itself can still be packaged as a single product file.
After installing the packaged release on another machine:
- Open IntentOps Studio.
- Confirm the splash screen appears.
- Create a new encrypted project.
- Close and reopen the application.
- Unlock the project.
- Import a Business Profile JSON.
- Import a Conversation Export JSON.
- Review and confirm samples.
- Build a dataset.
- Train a model.
- Open Evaluation.
- Open Recommendations.
- Generate Reports.
- Export packages.
- Confirm the app opens again after restart.
IntentOps Studio 1.0.0 is the first public release focused on the complete local workflow:
Import
Review
Dataset
Train
Evaluate
Recommend
Report
This version is designed to prove the full product loop and provide a practical offline workspace for intent operations.
Possible future improvements:
- larger dataset management tools;
- dataset merge workflow;
- richer report templates;
- model comparison dashboard;
- batch evaluation;
- active learning suggestions;
- advanced confusion analysis;
- direct chatbot platform integrations;
- additional import adapters;
- improved sample extraction heuristics;
- project templates;
- organization-level workspaces;
- richer visualization for intent drift and coverage gaps.
Initial public release.
Highlights:
- Added local project creation.
- Added optional encrypted local project database workflow.
- Added Business Profile import/export and editor.
- Added structured conversation import.
- Added sample extraction and review workflow.
- Added dataset builder.
- Added dataset export as JSON, JSONL, CSV, and package.
- Added duplicate dataset package protection.
- Added local model training.
- Added model family and training run workflow.
- Added model package export/import.
- Added duplicate model package protection.
- Added evaluation metrics and confusion matrix.
- Added recommendations and gaps.
- Added recommendations/gaps JSON export.
- Added reports generation and recent reports.
- Added project backup and restore.
- Added polished dark WPF interface.
- Added release-ready Windows installer workflow.
IntentOps Studio is released under the MIT License.
See:
LICENSE
Third-party libraries remain under their respective licenses.
Created by Micilini Roll.
- Website: https://micilini.com
- GitHub: https://github.com/micilini
IntentOps Studio was built to make intent operations practical, local, and understandable.