Models (/models) is the control plane for production machine learning in your organization — separate from the lighter Training fine-tune flow. GIBE (Gravitre Intelligent Business Engine) supplies the built-in ML catalog at Built-in models (/models/built-in): org-scoped rankers, predictors, and evaluators trained on verified signals in your org.
Requires Command plan or above for API access.
Model layers
The registry organizes work into layers:
| Layer | Typical use |
|---|---|
| Generative & agents | Fine-tuned LLMs, agent routing |
| Tabular & classification | Lead scoring, churn, routing |
| Forecasting | Demand and revenue projections |
| Anomaly detection | Ops and billing outliers |
| MLOps | Deployment pipelines tied to workflows |
| GIBE built-in | Pre-trained org models — browse at /models/built-in |
Templates suggest model type, task, and base model for each layer.
Lifecycle
Draft → Training / validating → Ready → Deployed → (archived)
Deployed models expose predict endpoints for workflows and services. Training may pull from connected Sources when configured. GIBE built-in models update as your org accumulates verified run and chat signals.
Models vs Training
| Training | Models | |
|---|---|---|
| Focus | Agent fine-tunes, instructions, quick jobs | Full registry, GIBE built-in catalog, deploy, inference |
| Audience | Builders improving agents | ML ops / advanced teams |
| Plan | Standard paid tiers | Command+ |
Many teams start in Training and promote stable models into Models for production deployment. GIBE built-in models are available without a custom training job once your org meets signal thresholds.
Next steps
- Manage Models
- GIBE — Learning for org memory and built-in model evaluation
- Connect Sources for training data