How to Build and Configure Boltis: Voice AI Agents in 2026

Bolti, a voice AI platform for phone agents, allows businesses to deploy human-like conversational assistants in minutes. Whether you are running an outbound sales campaign, managing customer support, or handling after-hours helpdesks, building these custom AI agents—often referred to by users as "Boltis"—requires just a few simple configuration steps. With Bolti's ₹6/min pay-as-you-go pricing and a 50-minute free trial, you can go from a blank canvas to a production-ready voice agent without writing complex code.

In this guide, we will walk you through how to set up, configure, and optimize your Boltis to handle real-world phone calls with sub-second latency and natural interruption handling.

What are Boltis and how do you configure them?

Boltis are conversational voice AI agents built on the Bolti platform that automate outbound and inbound phone calls. You configure them primarily through two high-leverage control panels in your dashboard: the Basic Tab and the LLM Tab.

While it is tempting to focus entirely on choosing the largest language model, the secret to a high-performing voice agent lies in how you define its identity, goals, and guardrails. The Basic tab is where you shape who the agent is and what it is trying to do. Most quality issues, such as going off-topic or failing to collect lead details, are fixed here rather than in the LLM settings.

Step 1: Define the core identity on the Basic Tab

To build a reliable voice agent, you must first establish its core parameters. Navigate to the Basic tab in your Bolti dashboard to configure these foundational settings:

• Agent Name & Greeting: Set a display name for dashboard tracking. You can toggle the first message (greeting) on or off. For inbound support, turn it on to greet callers immediately (e.g., "Hi, this is Aria from Acme Support. How can I help you today?"). For outbound calls, toggle it off so the agent waits for the customer to speak first.
• Persona Selection: Choose a high-level archetype that pre-tunes your system prompt. Bolti offers pre-built personas like Helpdesk (calm, patient troubleshooting), Sales (confident, consultative outbound), and Healthcare (empathetic intake and appointment confirmation).
• Primary Language & Timezone: Select the language your agent will speak and listen for. Bolti supports over 80 languages, including Hindi, Marathi, Tamil, Telugu, Bengali, Gujarati, and English. Selecting a language automatically updates the speech-to-text (STT) model for accurate transcription. Set the timezone (defaulting to Asia/Kolkata) to ensure time-of-day references like "good morning" are accurate.

Step 2: Set sharp goals and guardrails

Your voice agent needs a clear objective and strict boundaries to prevent it from hallucinating or sharing sensitive information during a call.

Writing an action-oriented goal

Your goal should be a single, specific sentence describing the exact outcome you want from the call. A specific goal produces dramatically better calls than a vague one.

• Vague (Avoid): "Talk to the customer about our software services."
• Specific (Recommended): "Qualify the lead by collecting budget, timeline, and decision-maker, then book a demo."

Setting up guardrails

Guardrails are forbidden topics that the agent must refuse to discuss. If a customer asks about a restricted topic, the agent will politely decline and steer the conversation back. Add short phrases as guardrails, such as:

• pricing for enterprise plans
• competitor product comparisons
• legal advice

Fetching company context

If your agent represents your business externally, enter your website domain (e.g., acme.com) and click Fetch Info. Bolti scrapes your homepage to extract a concise company-context paragraph and injects it directly into the system prompt. This gives your agent instant knowledge about your products and services.

Step 3: Choose the right LLM for voice performance

Once your agent's identity and prompt are defined, head to the LLM tab to select the underlying language model that will generate replies on every turn of the conversation.

For voice applications in 2026, three properties matter when choosing a model: first-token latency (aiming for under 400ms), instruction following, and cost per turn. You can choose from several top-tier providers:

Provider	Recommended Models	Best For
OpenAI	gpt-5-mini, gpt-5-nano, gpt-4o	Default starting point; balanced cost and speed
Gemini	gemini-2.0-flash-lite, gemini-1.5-pro	Low latency; excellent multilingual code-switching
Groq	llama-3.1-8b-instant	Lowest latency for ultra-fast response times
DeepSeek	deepseek-chat	Highly cost-effective conversational turns

By default, Bolti uses platform-managed credentials so you do not need to manage your own API keys. However, for enterprise compliance and lower costs at scale, you can easily bring your own keys (BYO).

Step 4: Add dynamic personalization and scheduling

To make your Boltis feel truly human, you can implement dynamic variables and automated scheduling features:

1. Dynamic Variables: Both the system prompt and the first greeting support dynamic variables using double curly braces {{ ... }}. For example, you can pass customer names or appointment details dynamically from your CRM: "Hello {{ name | default: "there" }}, your appointment is on {{ appointment_date }}." Learn more about customizing these on our Bolti use cases page.
2. Smart Call Scheduling: Toggle this on to enable automatic callback detection. If a customer says "call me tomorrow at 3 PM", Bolti's intent detection parses the time, registers an outbound task, and dials the customer back automatically at the requested time.

Once you have configured your settings, simply click Save Changes at the top of the panel to deploy your updated agent.

Set up your first Boltis agent today

You do not need a team of developers or complex infrastructure to deploy human-grade voice AI. With Bolti, you can build, test, and launch production-ready voice agents optimized for Indian languages and business contexts. Check out the Bolti pricing page to see our simple ₹6/min pay-as-you-go model.

Ready to see how voice AI can transform your customer operations? Spin up your first conversational voice agent in under 10 minutes—start your free trial with 50 free minutes today.

Frequently Asked Questions

What is the default LLM used for Boltis voice agents?

The default model for new agents is OpenAI's gpt-5-mini. It offers a highly optimized balance of fast first-token latency, strong instruction following, and low cost per conversational turn.

Can I use my own API keys with Bolti?

Yes. While Bolti provides platform-managed credentials by default so you can start calling immediately, you can easily bring your own API keys (BYO) for providers like OpenAI, Gemini, or DeepSeek to reduce costs at scale.

How do I prevent my voice agent from discussing certain topics?

You can set up Guardrails on the Basic tab. By adding short phrases of forbidden topics (e.g., 'competitor comparisons' or 'legal advice'), the agent is instructed to politely decline those topics and steer the conversation back to your primary goal.

Does Bolti support automatic callbacks?

Yes. By enabling Smart Call Scheduling on the Basic tab, the agent automatically detects callback requests (e.g., 'call me back in 10 minutes'), parses the requested time, and schedules an automatic outbound redial.