Managing Agents

What is an Agent?

An agent is your AI-powered receptionist that answers calls on your behalf. Unlike a traditional phone system, an agent can hold natural conversations, understand context, and follow complex instructions.

Each agent can be customized with:

• Voice — Choose from multiple AI voices to match your brand.
• Greeting — What the agent says when answering the phone.
• Language — The primary language for conversations.
• Business hours — When the agent is available to take calls.
• Knowledge — Documents and instructions about your business.
• Escalation rules — When and how to transfer calls to humans.

Creating a New Agent

1. Go to the Agents page

   Navigate to Dashboard → Agents (shown as "Team" in the sidebar).

2. Click "Hire New Agent"

   This opens the agent creation wizard.

3. Choose a name and purpose

   Give your agent a name (e.g., "Alex") and select what it's for: primary receptionist, dedicated line, agency client, or campaign.

4. Select a voice

   Preview different voices and choose one that fits your brand.

5. Customize the greeting

   If you completed onboarding, a smart greeting is generated automatically. You can edit it at any time.

6. Create the agent

   Your agent is now ready. Assign a phone number to start receiving calls.

Voice Selection

Infercall offers multiple AI voices powered by Google's latest speech technology. Each voice has unique characteristics suitable for different business types.

Customizing the Greeting

The greeting is the first thing callers hear. A good greeting identifies your business, feels warm, and clearly asks how the agent can help.

Example Greetings:

• "Good morning, thank you for calling Bella's Salon! I'm Alex, your AI agent. How can I help you today?"
• "Hi there! You've reached Downtown Auto Service. I'm here to help with appointments, quotes, or any questions. What can I do for you?"

The greeting is stored on the agent's Basic Settings tab and can be edited at any time.

Business Hours

Configure when your agent is available to take calls. Outside business hours, you can configure the agent to handle calls differently (e.g., take a message, provide hours, or forward to voicemail).

1. Open your agent's settings.
2. Go to the Business Hours tab.
3. Set open/close times for each day of the week.
4. Mark specific days as closed if needed.
5. Ensure your timezone is set correctly.

Advanced Settings

Agent Status

Set to Active to receive calls, or Inactive to temporarily disable the agent. Inactive agents won't answer calls on their assigned numbers.

Language

Choose the primary language for your agent. While the agent can understand multiple languages, it will primarily respond in the selected language unless prompted otherwise.

Temperature

Controls the creativity of the agent's responses.

• 0.3 - 0.5: More predictable, concise, and factual. Good for medical or legal contexts.
• 0.7 (Default): Balanced creativity and consistency.
• 0.8 - 1.0: More varied and conversational.

Caller Memory

When enabled, the agent remembers repeat callers and their preferences from previous conversations.
Note: This feature may incur additional API costs per call depending on your plan.

Agent Knowledge

Your agent draws knowledge from multiple sources to answer questions accurately:

• Website scraping — Information gathered from your website during onboarding.
• Business Info — Documents (PDF, DOCX, TXT) you upload.
• Personality — Advanced instructions on how your agent talks and behaves.

A2A Test Controller (Admin, Staging)

The A2A Test Controller helps operators run controlled agent-vs-agent (A2A) calls and verify clean handoff + hangup behavior before rolling prompt or model changes to production.

Who should use it

• Ops / QA validating changes to prompts, voice behavior, or turn-taking logic.
• Support reproducing edge cases where calls do not close cleanly.
• Engineering testing a specific receiver agent against a known caller harness.

Quick start

1. Open Dashboard → Admin → A2A Hangup.
2. Set callerAgentId (default holly-outbound-caller).
3. Set receiverAgentId to test any existing receiver agent. Leave it blank to use receiverPromptOverride as a synthetic receiver.
4. Optionally tune callerSteerPrompt and callerKickoffPrompt.
5. Save, then run npx tsx scripts/outbound/test-agent-vs-agent-vertex.ts.

Key fields

• Caller / Receiver IDs — Select the exact pair of agents you want to test.
• Steer + Kickoff prompts — Add temporary test directives to the caller and control the first utterance that starts the call.
• Hangup regex controls —callerEndIntentRegex, receiverAckRegex, and continuationRegexdefine when the harness should close vs keep talking.

Safe defaults

• Keep the default regex values unless you are fixing a specific false positive/negative.
• Use settleMs=1800 and fallbackSilenceMs=1800 as the baseline quiet window.
• Keep ackLagToleranceMs=700 to tolerate short transcript lag.
• Keep minCallMs=18000 so calls do not auto-close too early.
• Leave requireReceiverTurnComplete=false unless you are explicitly debugging turn-complete signaling.

Troubleshooting

• "Agent not found": verify the receiver/caller agent ID exists in staging Firestore.
• Receiver never responds: if receiverAgentId is blank, provide areceiverPromptOverride and optional greeting.
• Call hangs too long: reduce settleMs slightly or check ifcontinuationRegex is too broad.
• Call closes too early: increase minCallMs and reviewcallerEndIntentRegex for accidental matches.

Voicemail QA Console (Admin, Staging)

Use the Voicemail QA Console to verify the full fallback pipeline: transfer attempt → voicemail capture → notification target configuration → recording playback readiness.

1. Open Dashboard → Admin → Voicemail QA.
2. Optionally filter by org ID to focus on one customer environment.
3. Review pass/fail stage cards for each voicemail call.
4. Use Send test notification to validate email wiring independently.

Agent Limits by Plan