Custom Model Training

Train Phony to generate data that matches YOUR domain. Works with any language—Turkish, Japanese, or your custom terminology.

---

How It Works

Phony learns statistical patterns from your data, not memorizes it. The trained model generates new data that looks like yours but never reproduces original values.

Your Data → Learn Patterns → Generate Similar (Not Same)

• Learns character/word distributions
• Preserves statistical properties
• Never reproduces original data
• Works with ANY language

---

Training Sources

Open Source (Local)

• Text file (.txt, one item per line)
• CSV/TSV (any column)
• JSON/JSONL (any field path)

# CLI
phony train --source names.txt --output names.phony

# PHP
Phony::train($data, ['ngram' => 2]);

Cloud (Additional Sources)

• Database column (MySQL, PostgreSQL, SQLite)
• API response (fetch & extract)
• S3/GCS bucket (bulk data)
• Clipboard/Paste (quick training in Web UI)

---

Cloud Workflow

1. Connect database or upload file
2. Select column/field to train from
3. Configure n-gram size, excludeOriginals, etc.
4. Preview generated samples in UI
5. Train & Save model (with version)
6. Use in sync jobs or download for local use

---

Model Management

• List all models with stats
• Generate samples to test model quality
• Download .phony file for local/OSS use
• Share within your team
• Version history with diff
• Rollback to previous version
• Scheduled re-training from live data

---

Usage Examples

# Train locally
$ phony train --source turkish_names.txt -o turkish.phony

# Load and use
Phony::loadModel('turkish.phony');
Phony::model('turkish')->generate(1000);

# In Cloud sync jobs
first_name: phony:model:turkish_names_v1

---

Tier Limits

Feature	FREE	STARTER	TEAM	BUSINESS
Cloud Models	3	Unlimited	Unlimited	Unlimited
Database Column Training	✗	✓	✓	✓
S3/GCS Training	✗	✗	✓	✓
Model Versioning	✗	Last 5	Unlimited	With diff
Team Sharing	✗	✗	✓	✓
Scheduled Re-training	✗	✗	✓	✓