Train your agent

Ground the model

• Knowledge sources: Upload FAQs, product docs, and pricing sheets; or point the agent at a URL crawl or knowledge base to keep answers fresh.
• Tone & persona: Choose a voice profile (friendly, concise, expert), set greeting/opening/closing lines, and block phrases you never want used.
• Language & locale: Select supported languages/accents and default transcription settings for your callers.

Conversation design

• Intents and flows: Define primary intents (sales lead, support triage, scheduling, collections) and the sequence of prompts to reach an outcome.
• Data capture: Mark required fields (name, contact, need, budget, notes) and whether they’re blocking before the call/chat can close.
• Escalations: Configure triggers for handoff (customer asks for human, high-value lead detected, sentiment drops, time-on-call exceeded).

Testing & quality

• Dry-run scripts in the simulator and check slot-filling coverage.
• Place test calls and WhatsApp chats from multiple devices; confirm barge-in, silence handling, DTMF capture, and fallback behaviors.
• Review transcripts, redline incorrect responses, and add clarifying snippets to tighten answers.

Governance

• Enable transcript retention rules and PII redaction where needed.
• Set who can publish changes versus who can propose training updates.
• Document rollbacks so you can revert to the last stable prompt set quickly.

With training locked in, you’re ready to launch campaigns in /run-campaigns/.