Use Training

Open BUILD → Training (/training) to improve agent behavior with your org's examples and instructions.

Create a dataset

1. In Datasets, enter a name and choose a type (for example examples)
2. Click Create dataset
3. Open the dataset and Add records — each record has input text and expected output
4. Use starter examples in the UI as templates for support or ops scenarios

Wait until the dataset status shows ready before starting a training job.

Start a training job

1. Go to Training jobs
2. Select a ready dataset and base model (for example a supported GPT or Claude base)
3. Start the job and monitor status: queued → training → completed or failed
4. Cancel in-flight jobs from the job detail actions if needed

Completed jobs produce or link fine-tuned models you can assign to agents.

Custom instructions

Custom instructions are reusable prompt fragments scoped to the org or a specific agent:

1. Open Instructions
2. Create a name and body text
3. Toggle active when ready
4. Filter by agent when instructions target one worker

Instructions apply without running a full fine-tune — useful for policy, tone, and domain vocabulary.

Assign a fine-tuned model to an agent

1. Open Fine-tuned models or Workflow agents section
2. Choose an agent and select the completed model
3. Save assignment
4. Test the agent in chat or via a workflow step before production traffic

Monitor on Schedules

Active or scheduled training jobs may appear on Schedules with kind Job (green). Use that view alongside Runs for operational awareness.

API

# List datasets
curl -s "https://gravitre.app/api/training/datasets" \
  -H "Authorization: Bearer YOUR_API_KEY"

# Create dataset
curl -s -X POST "https://gravitre.app/api/training/datasets" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"name": "Support examples", "type": "examples"}'

# Start training job
curl -s -X POST "https://gravitre.app/api/training/jobs" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"dataset_id": "DATASET_ID", "base_model": "gpt-4.1-mini"}'

# List custom instructions
curl -s "https://gravitre.app/api/training/instructions" \
  -H "Authorization: Bearer YOUR_API_KEY"

See OpenAPI for record upload, cancel, and agent assignment endpoints.

Related

• Training overview
• Build with Meson
• Manage agents