Training (/training) is where teams curate examples, run fine-tuning jobs, and attach custom behavior to agents.
Main areas
| Area | Purpose |
|---|---|
| Datasets | Labeled input/output examples for fine-tuning |
| Training jobs | Track queued, running, and completed training runs |
| Custom instructions | Org- or agent-scoped instruction snippets |
| Fine-tuned models | Models produced from jobs, assignable to agents |
| Workflow agents | Link trained models to specific agents |
How it fits together
- Create a dataset (for example type
examples) and add records - Start a training job against a base model
- When complete, register or assign the fine-tuned model to an agent
- Use custom instructions for lighter-weight tuning without a full job
Training jobs also appear on the Schedules dashboard under the green Job kind when they have planned or active run windows.
Meson connection
Meson can propose datasets, instructions, and training plans as part of a system build. Review Meson output in Training before promoting to production agents.
Plans
Core dataset and instruction features are available on standard paid plans. Heavy fine-tuning volume may be metered — check Settings → Billing usage for your org.
Next steps
- Use Training
- Models for production ML registry, GIBE built-in catalog, and deployment