The Agents dashboard is where you design the core intelligence of your voice infrastructure. An Agni agent is more than just a prompt — it is a fully configurable voice persona with customizable speech patterns, tool access, and call handling behavior.
Agni Agents Dashboard

​
Quick Start

Get your first agent live in under 5 minutes:

Click + New Agent

From the Agents dashboard, click + New Agent to open the agent builder.

Write a system prompt

Describe who the agent is, what it does, and how it should handle calls. Be specific about tone, objectives, and edge cases.

Pick a voice and model

Select an LLM model and voice profile. Preview voices before committing.

Set a welcome message

Write the opening line the agent speaks when the call connects (e.g., “Hi, this is Ava from Acme. How can I help?”).

Test with a web call

Click Start Web Call in the testing panel to have a live conversation with your agent through your browser.

​
Creating a New Agent

When you click + New Agent, you are taken to the agent builder interface. This is where you define the agent’s behavior, knowledge, and capabilities.
New Agent Builder Interface

​
Prompt & Dynamic Variables

The main text editor is where you write the System Prompt, defining the agent’s role, objective, and conversational style. You can inject Dynamic Variables into your prompt using double curly braces (for example, {{user_name}} or {{company}}). During live calls, you can pass values into these variables via the API to personalize the conversation. When testing your agent in the dashboard, any variables you define will automatically generate an input field in the testing panel so you can simulate dynamic data.
Pro Tip: Use dynamic variables to pass user-specific context like CRM data or recent account activity directly into the agent’s working memory at the start of a call.
Example system prompt: 
You are {{agent_name}}, a sales assistant at {{company}}.
Your job is to qualify inbound leads by asking about their
budget, timeline, and use case. Be warm, professional, and
concise. If the caller is a good fit, offer to book a demo.
If not, thank them and end the call politely.

​
AI Provider Settings

Click the settings icon next to the model name to configure the underlying LLM:
  • Model: Choose the reasoning engine (e.g., Thunder Emotion Lite) powering the agent.
  • Voice: Click the voice selector button on the right side of the agent model field to open the voice picker. You can preview voices and select the one your agent should use.
  • Temperature: Adjust the creativity of the responses. Lower values (0.1—0.3) produce strict, predictable outputs; higher values (0.7—1.0) allow for more dynamic conversation.
  • Structured Output: Toggle this to force the model to return strict, schema-compliant responses (useful for data extraction).
AI Provider Settings

​
Welcome Message & Turn-Taking

At the bottom of the prompt editor, configure how the call begins:
  • Welcome Message: Define the exact phrase the AI should say when the call connects.
  • Start Speaker: Choose whether the User speaks first or the AI starts speaking after silence.
For outbound calls, the AI should speak first. For inbound calls, it’s common to let the user speak first or have the AI greet them.

​
Agent Configuration

The right sidebar contains granular controls for your agent’s capabilities and telephony behavior.
Equip your agent with custom tools. This allows the AI to fetch real-time data, trigger webhooks, or interact with external APIs during a conversation. See the Tools & Functions guide for details.
Enable scheduling capabilities. You can link your agent to a specific Lead Connector (GoHighLevel) calendar or a Cal.com event type, allowing the AI to negotiate times and book appointments automatically.
Calendar Integrations
You must first connect your GHL or Cal.com account in Workspace Settings before calendars will appear here.
Fine-tune the conversational dynamics and audio profile:
  • Background Sound: Add ambient noise (e.g., office hum, static) for realism.
  • Responsiveness & Interruption Sensitivity: Control how quickly the AI replies and how easily the user can interrupt it.
  • Enable Backchanneling: Allow the AI to use filler words (e.g., “Mhm”, “Yeah”) while listening.
  • Enable Speech Normalization: Convert complex numbers and dates into natural spoken forms.
  • Reminder Message Frequency: Configure how often the AI prompts a silent user.
Speech Settings
Configure the telephony rules for this specific agent:
  • Voicemail Detection: Automatically detect and leave a predefined message if a machine answers.
  • End call on silence: Disconnect automatically after a prolonged period of dead air.
  • Max duration: Set a hard limit on call length (useful for preventing runaway costs).
  • Post-call data extraction: Automatically summarize the call and extract key data points using an LLM.
Call Settings
Add webhook URLs to receive automatic event notifications (like transcripts, summaries, and extraction data) the moment a call concludes. This is how you push call data into your CRM, analytics tools, or custom backend.
Webhook Settings

​
Testing Your Agent

Agni provides a built-in testing environment so you can simulate conversations before deploying your agent to production.

​
Web Call

Test the agent directly through your browser’s microphone and speakers.
  • Input values for any Dynamic Variables you included in your prompt.
  • Inject custom Metadata (JSON) to simulate data passing.
  • Click Start Web Call to connect.
Test Web Call Interface

​
Phone Call

Test how the agent sounds over an actual cellular or VoIP network.
  • From Number: Select one of your provisioned Agni phone numbers to act as the caller ID.
  • To Number: Enter your personal cell phone number.
  • Click Start Phone Call to initiate an outbound call to your device.
Test Phone Call Interface
Testing tip: Try to break your agent. Ask unexpected questions, interrupt it mid-sentence, stay silent for long periods, and test all edge cases before going live.

​
Prompt Writing Best Practices

Don’t just say “You are a helpful assistant.” Specify the name, company, role, and personality. Example: “You are Sarah, a friendly customer support agent at CloudByte. You speak in short, clear sentences and always confirm details before proceeding.”
Tell the agent exactly what a successful call looks like. Is it booking a demo? Resolving a ticket? Collecting payment info? The AI needs a concrete goal to work toward.
Specify what to do when the caller is angry, confused, off-topic, or asks about competitors. Without guidance, the AI will improvise — and that’s usually not what you want.
Long, rambling prompts confuse the model. Use bullet points, short sentences, and clear structure. Under 500 words is ideal for most use cases.

​
API Reference

Once your agent is configured and tested, you can deploy it programmatically.

Create Agent

Create new agents programmatically via the REST API.

Update Agent

Update prompts, settings, and functions for existing agents.

List Agents

Retrieve all agents in your workspace.

Update Status

Activate or pause agents without changing their configuration.