Train Agent

Overview

The Train Agent endpoint queues a background job that re-ingests all knowledge sources linked to the agent and rebuilds the vector index. This is required after adding or removing knowledge sources, or after updating source content, to ensure the agent's responses reflect the latest information.

Path Parameters

Error Responses

404

{
    "error": {
        "code": "agent_not_found",
        "message": "Agent not found.",
        "status": 404
    }
}

400

{
    "error": {
        "code": "knowledge_sources_required",
        "message": "The agent must have at least one active knowledge source to start training.",
        "status": 400
    }
}

Side Effects

• Agent status is immediately set to "Training".
• All linked knowledge source statuses are set to "Training".
• On completion, agent status transitions to "Active" (success) or "Issues" (partial failure).

Best Practices

• Batch Changes First: Make all knowledge source additions and removals before triggering retrain to avoid multiple training cycles.
• Do Not Retrain Unnecessarily: Retraining is resource-intensive. Only trigger it when the knowledge base has actually changed.