Before you begin
You need access to a Valuez workspace and permission to create agents and knowledge bases. Prepare a short description of the use case, a sample greeting, five common customer questions, one unsupported question, and the destination for a human transfer. These items are enough for a useful first test. Write them down before opening the dashboard. Builders who prepare answers first finish faster because every required field - name, instructions, greeting, FAQs - draws from the same sheet.1
Choose a template
Open AI Voice Agents, select Create Agent, and choose Create from
Template or Create from Scratch, then Continue.
2
Enter agent details
Choose the name (30 chars), region, time zone, languages
(English/Hindi/Marathi + opening language), industry, agent type, tone,
icon, and instructions (10+ words, 1200 chars max).
3
Define its behavior
In Agent Settings → Agent Instructions and Conversation, describe the
agent’s role, supported tasks, tone, greeting and ending messages, fallback
response, and transfer number with country code.
4
Add knowledge
From Knowledge Base → Add Knowledge, connect approved business
information, FAQs (question 5+ chars, answer 10+ chars), products, services,
or PDF/Excel documents (10 MB max).
5
Choose voice settings
In LLM Model, Voice & Audio, Speech to Text, select providers and
models, response length 50–300 words and temperature, speaking speed
0.7–1.2, then Play preview and Save & Next.
6
Test the experience
In Test Agent, run Chat Test (consent + name/email) and Voice
Test (mute, interrupt, end call). Test a normal request, an unknown
question, an interruption, and a transfer request. Use Phone Call
Preview for a dry run without real calling.
Build the first version
Keep the initial scope narrow. For example, create an agent for appointment enquiries or product questions before combining both workflows. Give the agent a clear outcome, such as answering a question, collecting details, or transferring the caller. Add only current, approved information. If a document contains several unrelated subjects, split it into smaller sources with meaningful titles so answers are easier to verify.Core settings
Run a short test conversation
- Start with the expected greeting.
- Ask one question that is covered by the knowledge base.
- Correct a detail to confirm the agent follows the update.
- Ask an unsupported question and check the fallback response.
- Complete the intended action or request a transfer.
- Review the saved conversation record.
Confirm success
Your first agent is ready when it can answer an approved question, handle an unknown question without inventing information, understand expected callers, and save a complete conversation record. Use this exit checklist before connecting callers:- Greeting identifies the service and invites the caller to speak.
- One known question returns the approved answer worded naturally.
- One unknown question triggers the fallback without invented facts.
- One interruption and one correction are handled without losing context.
- Transfer reaches the correct destination or states a clear fallback.
- The conversation record shows transcript, summary, outcome, and duration.
What to do next
If any checklist item fails, fix it in this order: knowledge content first, agent instructions second, voice and transcription third, transfer and campaign setup last. Retest the failed case plus one passing case after each change so fixes do not break what already worked.Completing every setup field does not guarantee that the agent is ready for
production. Always complete the testing checklist.
Improve the prompt
Write clear instructions for reliable conversations.
Run final tests
Use the launch checklist before going live.