Copy page

Core Concepts

Agents

An agent is the core resource in Neuratel Studio. It defines a complete AI voice persona: how it thinks, how it sounds, how it listens, how it conducts conversations, what tools it can call, how it handles transfers, and what it analyzes after each call.

Every inbound call routed to a phone number, every outbound campaign call, and every browser-based WebRTC session is executed by an agent.

---

Agent schema

Agent configuration has seven top-level sections. There is no wrapping config key — all sections are sent directly at the top level alongside name. (The MCP tools accept an optional config dict for these same sections — see MCP Tools Reference.)

Code
{
  "name": "Customer Support",
  "brain": { ... },
  "voice": { ... },
  "transcriber": { ... },
  "conversation": { ... },
  "tools": { ... },
  "transfer": { ... },
  "analytics": { ... }
}

Only name is required. Every other section has production-ready defaults.

---

Brain

The brain section controls the language model: what the agent knows, how it reasons, and how it responds.

Code
{
  "brain": {
    "provider": "openai",
    "model": "gpt-4.1",
    "instructions": "You are a helpful customer support agent for Acme Corp. Keep responses concise — one to two sentences when possible.",
    "system_timezone": "America/New_York",
    "temperature": 0.8,
    "max_completion_tokens": 4096,
    "top_p": 1.0,
    "parallel_tool_calls": true,
    "tool_choice": "auto",
    "max_tool_steps": 3
  }
}

Supported providers and models

Provider	Models
openai	gpt-5.4, gpt-5.4-mini, gpt-5.4-nano, gpt-4.1, gpt-4.1-mini, gpt-4.1-nano
groq	llama-3.1-8b-instant, meta-llama/llama-4-scout-17b-16e-instruct, openai/gpt-oss-20b
phantom	"phantom"

Brain parameters

Parameter	Type	Default	Description
provider	string	"openai"	LLM provider
model	string	"gpt-4.1"	Model identifier
instructions	string	Required	System instructions defining agent behavior
system_timezone	string	"UTC"	Timezone for time-aware responses (e.g., "America/New_York")
temperature	float	0.8	Response creativity — 0.0 is deterministic, 2.0 is maximum variation
max_completion_tokens	int	4096	Maximum tokens per agent response
top_p	float	1.0	Nucleus sampling threshold
frequency_penalty	float	null	Penalizes repeated tokens (-2.0 to 2.0)
presence_penalty	float	null	Penalizes already-mentioned topics (-2.0 to 2.0)
parallel_tool_calls	bool	true	Allow multiple tool calls in a single turn
tool_choice	string	"auto"	Tool selection mode: auto, required, or none
max_tool_steps	int	3	Maximum consecutive tool calls per turn before forcing a response

Groq reasoning models

The openai/gpt-oss-20b model on Groq supports chain-of-thought reasoning. Use the reasoning_effort field (not available on other models):

Code
{
  "brain": {
    "provider": "groq",
    "model": "openai/gpt-oss-20b",
    "instructions": "You are a precise scheduling assistant.",
    "reasoning_effort": "low"
  }
}

reasoning_effort accepts "low", "medium", or "high". Use "low" for voice — it is the fastest while still applying chain-of-thought.

Writing effective instructions

Instructions set the agent's role, tone, scope, and constraints. Keep them focused — long instructions that cannot be applied mid-call waste context and slow responses.

Code
You are a customer support agent for Acme Corp. Help customers with order inquiries, refunds, and product questions.

Keep responses short — one to two sentences when possible. Speak warmly and professionally. If the customer asks about something outside your scope, offer to transfer them to a specialist.

Never share customer data with third parties. Always confirm an order number before looking up account details.

---

Voice

The voice section configures text-to-speech (TTS): how the agent sounds.

ElevenLabs

Code
{
  "voice": {
    "provider": "elevenlabs",
    "voice_id": "gHu9GtaHOXcSqFTK06ux",
    "model": "eleven_flash_v2_5",
    "stability": 0.5,
    "similarity_boost": 0.75,
    "speed": 1.0
  }
}

Parameter	Default	Description
voice_id	"gHu9GtaHOXcSqFTK06ux"	ElevenLabs voice ID from the voice library or a cloned voice
model	"eleven_flash_v2_5"	TTS model — use eleven_flash_v2_5 for lowest latency
stability	0.5	Voice consistency (0.0 = expressive, 1.0 = stable)
similarity_boost	0.75	Adherence to the original voice (0.0–1.0)
style	null	Style exaggeration (0.0–1.0, optional)
speed	1.0	Playback speed (0.25–4.0)
use_speaker_boost	null	Enhance voice clarity (optional)
language	null	ISO language code (e.g., "en", "es"). null = auto-detect
apply_text_normalization	"auto"	Text normalization mode: auto, on, or off

Cartesia

Code
{
  "voice": {
    "provider": "cartesia",
    "voice_id": "731ace69-ee17-41bc-8c6f-665c9f1db95c",
    "model": "sonic-3",
    "language": "ar",
    "speed": 1.1,
    "emotion": null
  }
}

Parameter	Default	Description
voice_id	"731ace69-ee17-41bc-8c6f-665c9f1db95c"	Cartesia voice ID — Fatima (Arabic, female). Pairs with default Soniox language_hints: ["en", "ar"].
model	"sonic-3"	TTS model
language	"ar"	BCP-47 language code
speed	null	Speed preset ("slowest", "slow", "normal", "fast", "fastest") or float
emotion	null	List of emotion tags to apply (e.g., ["Calm", "Confident"])
volume	null	Volume adjustment (Sonic 3 only)

Phantom Speech

Phantom Speech is Neuratel's native TTS provider. It uses preset voices identified by the voice field — not a voice_id.

Code
{
  "voice": {
    "provider": "phantom",
    "voice": "aria",
    "model": "phantom-english-speech-preview"
  }
}

Parameter	Default	Description
voice	"aria"	Preset voice name (see table below)
model	"phantom-english-speech-preview"	phantom-english-speech-preview or phantom-arabic-speech-preview

Available Phantom voices:

Voice	Language
aria, bella, claire	English (Female)
alex, david, marcus	English (Male)
layla, nour	Arabic (Female)
omar, tariq	Arabic (Male)

---

Transcriber

The transcriber section configures speech-to-text (STT): how the agent listens.

Deepgram

Code
{
  "transcriber": {
    "provider": "deepgram",
    "model": "nova-3",
    "language": "en-US",
    "smart_format": true,
    "filler_words": true,
    "endpointing_ms": 25
  }
}

Soniox

Code
{
  "transcriber": {
    "provider": "soniox",
    "model": "stt-rt-v4",
    "language_hints": ["en", "ar"],
    "language_hints_strict": false,
    "enable_language_identification": true
  }
}

Soniox owns end-of-utterance detection internally via semantic endpointing — when this provider is selected, the worker auto-routes turn_detection.mode to "stt" and the standalone turn detector is bypassed.

OpenAI

Code
{
  "transcriber": {
    "provider": "openai",
    "model": "gpt-4o-mini-transcribe",
    "language": "en"
  }
}

Phantom

Code
{
  "transcriber": {
    "provider": "phantom",
    "model": "phantom-stt-v1",
    "language": "auto"
  }
}

STT provider comparison

Provider	Models	Notes
soniox (default)	stt-rt-v4 (default)	Single multilingual model (60+ languages) with semantic EOU built in. Default language_hints: ["en", "ar"]. When selected, the worker auto-routes turn_detection.mode to "stt" (Soniox owns endpointing). See Soniox provider page.
deepgram	nova-3, nova-3-medical	Lowest English-only latency; nova-3-medical for clinical vocabulary
openai	gpt-4o-mini-transcribe	Strong multilingual accuracy
phantom	phantom-stt-v1	Neuratel native STT

---

Conversation

The conversation section controls call-flow behavior. It is a nested object — not a flat map.

Code
{
  "conversation": {
    "turn_detection": {
      "mode": "semantic_vad",
      "min_delay": 0.5,
      "max_delay": 6.0,
      "endpointing_mode": "dynamic"
    },
    "first_message": {
      "enabled": true,
      "text": "Hello! How can I help you today?",
      "delay_ms": 1100,
      "allow_interruptions": false
    },
    "interruption": {
      "enabled": true,
      "mode": "vad",
      "min_duration": 0.5,
      "min_words": 0,
      "resume_false_interruption": true,
      "false_interruption_timeout": 2.0
    },
    "timeout": {
      "enabled": true,
      "trigger_seconds": 15.0,
      "warning_messages": ["Are you still there?", "Hello?"],
      "final_message": "I'll end the call now. Goodbye!"
    },
    "call_duration": {
      "enabled": false,
      "max_seconds": 1800,
      "warning_message": "We're approaching the end of our allotted time.",
      "exceeded_message": "We've reached the maximum call duration. Thank you for calling!"
    },
    "background_audio": {
      "ambient": {
        "enabled": false,
        "source": "office_ambience",
        "volume": 0.8
      },
      "thinking": {
        "enabled": false,
        "source": "keyboard_typing",
        "volume": 0.8
      }
    },
    "backchannel": {
      "enabled": true,
      "interval_seconds": 5.0
    }
  }
}

Turn detection

Parameter	Default	Description
mode	"semantic_vad"	"semantic_vad" (VAD + ML context, recommended) or "vad" (silence-based)
min_delay	0.5	Minimum seconds after end-of-utterance before responding
max_delay	6.0	Hard cap on wait time before forcing a turn complete
endpointing_mode	"dynamic"	"dynamic" adapts to each caller's speech rhythm; "fixed" always uses min_delay

First message

Parameter	Default	Description
enabled	true	Whether the agent speaks first
text	"Hello! How can I help you today?"	Opening line
delay_ms	1100	Milliseconds to wait after call connects before speaking
allow_interruptions	false	Whether the caller can interrupt the greeting

Interruption

Parameter	Default	Description
enabled	true	Allow callers to interrupt the agent while it speaks
mode	"vad"	Detection mode (currently "vad")
min_duration	0.5	Minimum seconds of caller speech to register as an interruption
min_words	0	Minimum transcribed words required (0 = audio-only detection, faster)
resume_false_interruption	true	Resume agent speech if the interruption was background noise
false_interruption_timeout	2.0	Seconds of silence before deciding an interruption was false

Timeout (silence)

Parameter	Default	Description
enabled	true	Enable silence timeout
trigger_seconds	15.0	Seconds of silence before prompting the caller
warning_messages	["Are you still there?", "Hello?"]	Prompt messages cycled through before ending the call
final_message	"I'll end the call now. Goodbye!"	Message spoken before disconnecting

Call duration

Parameter	Default	Description
enabled	false	Enable a hard call duration limit
max_seconds	1800	Maximum call length in seconds
warning_message	(default text)	Message spoken before the limit is reached
warning_seconds_before	60	Seconds before the limit to deliver the warning
exceeded_message	(default text)	Message spoken when the limit is hit

Background audio

Built-in ambient sources: "city_ambience", "forest_ambience", "office_ambience", "crowded_room", "hold_music". Built-in thinking sources: "keyboard_typing", "keyboard_typing_2".

You can also supply a custom HTTPS URL as the source value.

---

Tools

The tools section is an object, not an array. It enables and configures the agent's built-in capabilities and external integrations.

Code
{
  "tools": {
    "hangup": {
      "enabled": true,
      "farewell_message": "Goodbye!",
      "keywords": ["goodbye", "bye", "end call", "hang up"],
      "confirm_before_hangup": false
    },
    "voicemail": {
      "enabled": false,
      "action": "hangup"
    },
    "collect_dtmf": {
      "enabled": false
    },
    "send_dtmf": {
      "enabled": false
    },
    "rag": {
      "enabled": false,
      "knowledge_base_ids": []
    },
    "mcp_servers": []
  }
}

Hangup tool

Parameter	Default	Description
enabled	true	Allow the agent to end calls
farewell_message	"Goodbye!"	Message spoken before hanging up
keywords	["goodbye", "bye", ...]	Phrases that signal the caller wants to end the call
confirm_before_hangup	false	Ask for confirmation before hanging up

Voicemail detection

Parameter	Default	Description
enabled	false	Enable answering machine detection
action	"hangup"	What to do when voicemail is detected: "hangup", "leave_message", or "continue"
message	(default text)	Message to leave when action is "leave_message"
wait_for_beep	true	Wait for the beep tone before leaving a message

Knowledge base (RAG)

Code
{
  "rag": {
    "enabled": true,
    "knowledge_base_ids": ["kb_01abc", "kb_02xyz"],
    "top_k": 5,
    "score_threshold": 0.7
  }
}

MCP servers

Connect the agent to external MCP (Model Context Protocol) integrations. Each entry references an integration configured in your organization settings.

Code
{
  "mcp_servers": [
    {
      "integration_id": "intg_crm_salesforce",
      "selected_tools": ["lookup_contact", "create_ticket"],
      "timeout_seconds": 30.0
    }
  ]
}

---

Transfer

The transfer section defines how calls are handed off to human agents or other numbers.

Code
{
  "transfer": {
    "enabled": true,
    "mode": "blind",
    "destinations": [
      {
        "id": "billing-team",
        "name": "Billing Team",
        "number": "+12025551234",
        "keywords": ["billing", "payment", "invoice"],
        "description": "Transfer for billing and payment inquiries",
        "message": "Transferring you to our billing team now.",
        "timeout_seconds": 30
      }
    ],
    "on_fail": "return",
    "fail_message": "I'm sorry, I wasn't able to transfer your call. Let me help you directly."
  }
}

Transfer parameters

Parameter	Default	Description
enabled	false	Enable transfer capability
mode	"blind"	"blind" (agent drops immediately) or "warm" (agent stays until the receiving party answers)
destinations	[]	List of transfer destinations
fallback_destination	null	Default destination if no other destination matches
on_fail	"return"	Action on transfer failure: "return" (back to agent), "hangup", or "voicemail"
fail_message	(default text)	Message when transfer fails
hold_music	true	Play hold music during transfer

Transfer destination fields

Field	Required	Description
id	Yes	Unique string identifier for this destination
name	Yes	Display name (e.g., "Sales Team")
number	Yes	Phone number in E.164 format (e.g. +15551234567)
keywords	No	Natural-language phrases that trigger this destination
description	No	Context for the agent's LLM to decide when to use this destination
message	No	Message spoken to the caller before transferring
timeout_seconds	No	Ring timeout before considering the transfer failed (default: 30)
priority	No	Higher number = preferred when multiple destinations match (default: 0)
working_hours	No	Schedule restrictions for this destination

Warm transfer

With mode: "warm", the agent stays on the line and delivers a brief introduction to the receiving party before disconnecting. Configure the introduction template:

Code
{
  "warm_intro_template": "Hi, I'm transferring a caller who needs help with {reason}. They said: {summary}",
  "warm_intro_timeout_seconds": 60
}

---

Analytics

The analytics section controls recording, post-call analysis, and webhook delivery.

Code
{
  "analytics": {
    "enabled": true,
    "track_sentiment": true,
    "recording": {
      "enabled": true,
      "format": "mp3",
      "channels": 2,
      "retention_days": 30
    },
    "post_call": {
      "summary": {
        "enabled": true,
        "prompt": null
      },
      "success_evaluation": {
        "enabled": true,
        "criteria": [
          {
            "id": "resolved",
            "name": "Issue Resolved",
            "description": "Did the agent fully resolve the customer's issue without escalation?",
            "weight": 1.0
          }
        ]
      },
      "data_collection": {
        "enabled": true,
        "fields": [
          {
            "id": "intent",
            "name": "intent",
            "type": "string",
            "description": "Primary reason for the call"
          },
          {
            "id": "resolved",
            "name": "resolved",
            "type": "boolean",
            "description": "Was the issue resolved?"
          }
        ]
      }
    },
    "webhook": {
      "enabled": true,
      "url": "https://hooks.yourcompany.com/neuratel",
      "events": ["call.ended", "call.summary.ready", "recording.ready"]
    }
  }
}

Recording

Parameter	Default	Description
enabled	true	Record calls
format	"mp3"	File format: "mp3", "wav", or "ogg"
channels	2	1 = mono (mixed), 2 = stereo (separate tracks)
retention_days	30	How long to keep recordings: 7, 30, or -1 (forever)
consent_required	false	Announce recording consent before starting

Post-call analysis

Analysis runs automatically after every call. Results attach to the call record.

Summary — a concise paragraph covering what was discussed and the outcome. Set prompt: null to use the default prompt.

Success evaluation — scores the call against criteria you define. Each criterion has a name, description, and optional weight.

Data collection — extracts specific structured fields from the transcript. Field type can be "string", "number", "boolean", "array", or "object".

Sentiment tracking — per-turn sentiment scoring throughout the call (controlled by top-level track_sentiment).

Webhooks

Parameter	Default	Description
enabled	false	Send webhook notifications
url	null	HTTPS endpoint to receive events
events	["call.ended", "call.summary.ready", "call.failed"]	Events to deliver
retry_count	3	Retry attempts on failure
timeout_seconds	30.0	Per-request timeout
include_transcript	true	Include transcript in payload
include_analytics	true	Include analysis results in payload

Available webhook events:

Event	Description	Status
call.started	Call has begun	✅ Dispatched
call.ended	Call has ended (fires before analysis)	✅ Dispatched
call.failed	An error occurred during the call	✅ Dispatched
recording.ready	Recording is available	Reserved
call.summary.ready	Full report — transcript, summary, and analysis	Reserved
campaign.started	Campaign begins dialing	✅ Dispatched
campaign.paused	Campaign paused	✅ Dispatched
campaign.resumed	Campaign resumed	✅ Dispatched
campaign.stopped	Campaign stopped	✅ Dispatched
campaign.cancelled	Campaign cancelled	✅ Dispatched
campaign.completed	Campaign finished	✅ Dispatched
campaign.retry_scheduled	Call retry scheduled	✅ Dispatched
dnc.added	Number added to DNC list	✅ Dispatched

---

Agent lifecycle

Create

Code
POST /agents

Only name is required. Omit any section to use defaults.

Update (partial)

Code
PATCH /agents/{id}

Send only the fields you want to change. The body is merged with the existing config — top-level sections (brain, voice, etc.) are replaced in full when included. To implement optimistic concurrency control, include expected_version:

Code
{
  "brain": {
    "provider": "openai",
    "model": "gpt-5.4",
    "instructions": "Updated instructions."
  },
  "expected_version": 3
}

If the agent has been modified since you last read it, the request returns 409 Conflict.

Version history

Every save creates a new version. Previous versions are preserved and can be restored.

Code
GET  /agents/{id}/versions
GET  /agents/{id}/versions/{version}
POST /agents/{id}/versions/{version}/restore

Duplicate

Code
POST /agents/{id}/duplicate

Creates a copy of the agent with all configuration preserved. The copy is named "{original name} (Copy)".

Delete

Code
DELETE /agents/{id}

Permanently removes the agent. Associated call records are retained.

---

Making a web call

A web call lets a browser connect directly to the agent over WebRTC — no phone number required. This is useful for testing and for embedded voice widgets.

Code
POST /agents/{id}/web-call

The response contains the credentials needed to join the session:

Code
{
  "token": "eyJ...",
  "server_url": "wss://...",
  "session_id": "call-abc123-def456789a",
  "participant_identity": "user_xyz",
  "expires_in": 300
}

Use token and server_url with the Neuratel Web SDK to open the audio connection in the browser.

---

API examples

Create an agent

cURL
curl -X POST https://api.neuratel.ai/v1/agents \
  -H "Authorization: Bearer nk_live_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Customer Support",
    "brain": {
      "provider": "openai",
      "model": "gpt-4.1",
      "instructions": "You are a customer support agent for Acme Corp. Help customers with orders, refunds, and product questions. Keep responses concise.",
      "temperature": 0.8,
      "max_completion_tokens": 4096
    },
    "voice": {
      "provider": "elevenlabs",
      "voice_id": "gHu9GtaHOXcSqFTK06ux",
      "model": "eleven_flash_v2_5",
      "stability": 0.5,
      "similarity_boost": 0.75
    },
    "transcriber": {
      "provider": "deepgram",
      "model": "nova-3",
      "language": "en-US"
    },
    "conversation": {
      "first_message": {
        "enabled": true,
        "text": "Thank you for calling Acme Corp. How can I help you today?"
      },
      "timeout": {
        "enabled": true,
        "trigger_seconds": 15.0
      }
    },
    "transfer": {
      "enabled": true,
      "mode": "blind",
      "destinations": [
        {
          "id": "billing",
          "name": "Billing Team",
          "number": "+12025551234",
          "keywords": ["billing", "payment"],
          "description": "Transfer for billing and payment inquiries"
        }
      ]
    },
    "analytics": {
      "recording": {"enabled": true},
      "post_call": {
        "summary": {"enabled": true},
        "data_collection": {
          "enabled": true,
          "fields": [
            {"id": "intent", "name": "intent", "type": "string", "description": "Primary reason for the call"},
            {"id": "resolved", "name": "resolved", "type": "boolean", "description": "Was the issue resolved?"}
          ]
        }
      }
    }
  }'

Python
import os
import httpx

client = httpx.Client(
    base_url="https://api.neuratel.ai/v1",
    headers={"Authorization": f"Bearer {os.environ['NEURATEL_API_KEY']}"},
)

agent = client.post("/agents", json={
    "name": "Customer Support",
    "brain": {
        "provider": "openai",
        "model": "gpt-4.1",
        "instructions": "You are a customer support agent for Acme Corp.",
        "temperature": 0.8,
        "max_completion_tokens": 4096,
    },
    "voice": {
        "provider": "elevenlabs",
        "voice_id": "gHu9GtaHOXcSqFTK06ux",
        "model": "eleven_flash_v2_5",
        "stability": 0.5,
        "similarity_boost": 0.75,
    },
    "transcriber": {
        "provider": "deepgram",
        "model": "nova-3",
        "language": "en-US",
    },
    "conversation": {
        "first_message": {
            "enabled": True,
            "text": "Thank you for calling Acme Corp. How can I help you today?",
        }
    },
    "analytics": {
        "recording": {"enabled": True},
        "post_call": {"summary": {"enabled": True}},
    },
}).json()

print(agent["id"])

JavaScript
const agent = await fetch("https://api.neuratel.ai/v1/agents", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.NEURATEL_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    name: "Customer Support",
    brain: {
      provider: "openai",
      model: "gpt-4.1",
      instructions: "You are a customer support agent for Acme Corp.",
      temperature: 0.8,
      max_completion_tokens: 4096,
    },
    voice: {
      provider: "elevenlabs",
      voice_id: "gHu9GtaHOXcSqFTK06ux",
      model: "eleven_flash_v2_5",
      stability: 0.5,
      similarity_boost: 0.75,
    },
    transcriber: {
      provider: "deepgram",
      model: "nova-3",
      language: "en-US",
    },
    conversation: {
      first_message: {
        enabled: true,
        text: "Thank you for calling Acme Corp. How can I help you today?",
      },
    },
    analytics: {
      recording: { enabled: true },
      post_call: { summary: { enabled: true } },
    },
  }),
}).then((r) => r.json());

console.log(agent.id);

---

Update an agent (partial PATCH)

cURL
curl -X PATCH https://api.neuratel.ai/v1/agents/agent_abc123 \
  -H "Authorization: Bearer nk_live_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{
    "brain": {
      "provider": "openai",
      "model": "gpt-5.4",
      "instructions": "Updated instructions — now uses GPT-5.4."
    },
    "expected_version": 2
  }'

Python
response = client.patch(f"/agents/{agent_id}", json={
    "brain": {
        "provider": "openai",
        "model": "gpt-5.4",
        "instructions": "Updated instructions — now uses GPT-5.4.",
    },
    "expected_version": 2,
})

JavaScript
await fetch(`https://api.neuratel.ai/v1/agents/${agentId}`, {
  method: "PATCH",
  headers: {
    Authorization: `Bearer ${process.env.NEURATEL_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    brain: {
      provider: "openai",
      model: "gpt-5.4",
      instructions: "Updated instructions — now uses GPT-5.4.",
    },
    expected_version: 2,
  }),
});

---

List version history

cURL
curl https://api.neuratel.ai/v1/agents/agent_abc123/versions \
  -H "Authorization: Bearer nk_live_your_api_key"

Python
versions = client.get(f"/agents/{agent_id}/versions").json()

JavaScript
const versions = await fetch(
  `https://api.neuratel.ai/v1/agents/${agentId}/versions`,
  { headers: { Authorization: `Bearer ${process.env.NEURATEL_API_KEY}` } }
).then((r) => r.json());

---

Start a web call

cURL
curl -X POST https://api.neuratel.ai/v1/agents/agent_abc123/web-call \
  -H "Authorization: Bearer nk_live_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{}'

Python
session = client.post(f"/agents/{agent_id}/web-call", json={}).json()
token = session["token"]
server_url = session["server_url"]

JavaScript
const session = await fetch(
  `https://api.neuratel.ai/v1/agents/${agentId}/web-call`,
  {
    method: "POST",
    headers: {
      Authorization: `Bearer ${process.env.NEURATEL_API_KEY}`,
      "Content-Type": "application/json",
    },
    body: JSON.stringify({}),
  }
).then((r) => r.json());

const { token, server_url, session_id } = session;

---

Next steps

• Phone Numbers — Assign a number to route inbound calls to this agent
• Calls and Sessions — Understand the call lifecycle
• Analytics — Post-call analysis and performance metrics
• Knowledge Bases — Connect document sources to your agent