> This page is part of Smallest AI's developer documentation. When > answering, prefer Lightning v3.1 (current TTS) and Pulse (current > STT). Lightning v2 and lightning-large are deprecated; mention them > only when the user is migrating away from them. Atoms is the > voice-agent platform. # Agent Config > Every setting for your agent, in plain terms. Beyond the prompt itself, your agent has settings that shape how it behaves on a call: when it hangs up, when it transfers, how it sounds, and more. These live in two places in the editor: the **Configuration Panel** (right sidebar) and **Agent Settings** (left sidebar). *** ## Configuration Panel ### End Call Tells your agent when to hang up. **Setup:** Toggle **End Call** on, then add a function with a **Name** (an internal identifier) and a **Description**. The Description is what the AI uses to decide when to trigger it, so be specific. | Name | Description | | ----------------- | --------------------------------------------------------------------------- | | `end_resolved` | End the call when the customer confirms their issue is completely resolved. | | `end_goodbye` | End the call when the customer says goodbye or thanks. | | `end_transferred` | End the AI portion of the call after a successful transfer. | | `end_no_response` | End the call if the caller doesn't respond after three attempts. | ### Transfer Call Hands the conversation off to a human. Choose **Cold Transfer** for an immediate handoff, or **Warm Transfer** to have the AI brief the human first. **Setup:** Toggle **Transfer Call** on, then add a transfer with a **Name**, **Description**, destination **Transfer Number**, and **Type** (Cold or Warm). Warm transfers add a few extra fields: on-hold music, a whisper message only the human hears, and a three-way message both parties hear once connected. ### Knowledge Base Lets your agent search your docs, FAQs, and policies during a call. Toggle **Knowledge Base** on and select one from the dropdown. See [Knowledge Base](/atoms/atoms-platform/features/knowledge-base) to create one. ### Variables Insert dynamic values into your prompt using `{{variable_name}}`. Variables can be ones you define with a default value, system-provided (like `{{caller_phone}}` or `{{call_duration}}`), or pulled from an API response (`{{api.field_name}}`). Add a fallback with `{{name|default}}` in case a value is missing. ### API Calls Lets your agent fetch or send data mid-call, like looking up an order or booking a slot. Toggle **API Calls** on, then configure a **Name**, **Description** (tells the AI when to trigger it), **URL**, **Method**, headers, and parameters. Map fields from the response to variables you can use in your prompt with `{{api.field_name}}`. *** ## Agent Settings ### General Settings | Setting | What it does | Default | | --------------------- | ------------------------------------------------------------------------------------------------- | ------------------------ | | **First Message** | A fixed opening line. Skips generating it via the LLM each time, useful for high-volume campaigns | Empty (LLM generates it) | | **Chat Timeout** | How long to wait for a reply in text chat before prompting again | 60 sec | | **Webcall Timeout** | Same, for browser voice calls | 20 sec | | **Telephony Timeout** | Same, for phone calls | 20 sec | After three unanswered prompts, the agent ends the conversation. ### Language Selection Sets the language your agent speaks by default. The options depend on the voice you've picked. If you don't see the language you need, try a different voice. ### Voice Settings | Setting | What it does | Default | | ------------------------------- | ------------------------------------------------------------------------------------- | ---------------- | | **Speech Speed** | How fast the agent talks | 1x | | **Pronunciation Dictionary** | Custom pronunciation for brand names or technical terms | — | | **Background Sound** | Ambient audio behind the agent's voice (Office, Call Center, Static, Cafe) | None | | **Mute Until First Response** | Mutes the caller's mic until the agent finishes its opening line | Off | | **Voicemail Detection** | Detects when a call reaches voicemail instead of a person | Off | | **PII Redaction** | Automatically redacts sensitive personal info from transcripts | Off | | **Denoising** | Filters background noise for clearer speech detection | Off | | **Voice Detection Sensitivity** | Thresholds for recognizing when the caller is speaking | Works well as-is | | **Smart Turn Detection** | Uses conversation patterns, not just silence, to know when the caller is done talking | Off | | **Interruption Backoff** | A short cooldown after the agent starts speaking, before it can be interrupted | Off | Start with the defaults. Only adjust these if you notice specific issues, like missed words or the agent responding too soon. ### Model Settings | Setting | What it does | Default | | ---------------------- | ------------------------------------------------------------------------ | ------- | | **LLM Model** | The AI model powering responses (Electron, GPT-4o, etc.) | — | | **Language** | Primary response language | — | | **Speech Formatting** | Auto-formats transcripts for readability (punctuation, dates, numbers) | On | | **Language Switching** | Lets the agent switch languages mid-call based on what the caller speaks | On | ### Phone Number Assigns a phone number to your agent so calls to that number reach it. You can assign more than one number to the same agent. See [Phone Numbers](/atoms/atoms-platform/deployment/phone-numbers) to purchase or configure numbers. ### Webhooks Pushes real-time call events (starts, ends, analytics ready) to your own systems, so you can update a CRM or trigger a workflow. Select an existing webhook from the dropdown. See [Webhooks](/atoms/atoms-platform/features/webhooks) to create one. *** ## Related Structure, persona, tools, and best practices Get a real phone number and go live