Copy page

Getting Started

Platform Overview

Neuratel Studio connects AI agents to phone networks and gives you APIs and a dashboard to manage everything in between. This page explains how the core resources relate to each other and how calls flow through the system.

Core Resources

Resource	What It Is	Key Actions
Agent	An AI voice agent with its own brain (LLM), voice (TTS), transcriber (STT), conversation rules, tools, transfer rules, and analytics config	Create, configure, test, duplicate, activate/deactivate
Phone Number	A number assigned to your organization, linked to an agent	Assign agent, configure routing
Call	A single inbound or outbound phone conversation handled by an agent	View transcript, play recording, see analytics
Campaign	An automated outbound calling job — contact list + agent + schedule	Create, start, pause, stop, monitor progress
Workflow	A visual DAG (directed acyclic graph) of conversation nodes that drives branching logic	Design in Studio, link to a workflow agent
Knowledge Base	Documents and data your agent can reference during a call to answer questions accurately	Upload files, add text entries, link to agents
Webhook	An HTTP endpoint that receives notifications when call events happen	Subscribe to events, verify signatures

How It All Fits Together

Code
                    ┌─────────────────────────────────────────┐
                    │               Dashboard                 │
                    │  Agents · Campaigns · Analytics · Team  │
                    └─────────────┬───────────────────────────┘
                                  │ session auth
                    ┌─────────────▼───────────────────────────┐
                    │           Platform API                   │
   API keys ───────►│  /v1/agents  /v1/voice-sessions  /v1/voice-sessions/outbound    │
                    │  /v1/campaigns  /v1/webhooks  ...        │
                    └─────────────┬───────────────────────────┘
                                  │
              ┌───────────────────┼───────────────────┐
              ▼                   ▼                   ▼
     ┌────────────────┐  ┌───────────────┐  ┌────────────────┐
     │ Phone Numbers  │  │  AI Worker    │  │   Webhooks     │
     │ SIP provider   │  │  STT→LLM→TTS │  │   Your server  │
     └────────┬───────┘  └───────────────┘  └────────────────┘
              │
         PSTN / VoIP

The flow:

1. Your API or dashboard creates resources (agents, campaigns, webhooks)
2. Phone numbers route inbound calls to agents, or the API triggers outbound calls
3. The AI worker runs the conversation — speech recognition, language model, speech synthesis
4. When calls start, end, or transfer, webhooks fire to your server
5. Call records with transcripts and analytics are available via the API and dashboard

Agent Configuration Schema

Every agent is built from seven sections:

Section	Required	Purpose
brain	Yes	LLM provider, model, system prompt, temperature, and reasoning settings
voice	No*	TTS provider, voice selection, stability, speed, emotion controls
transcriber	No*	STT provider, model, language, formatting options
conversation	No	First message, turn detection mode, silence timeout, interruption behavior
tools	No	MCP server connections, knowledge base links, custom function tools
transfer	No	Rules for transferring to a human or another agent (by keyword, intent, or tool call)
analytics	No	Custom metrics to extract, webhook destinations for analytics events

*Voice and transcriber are not needed when using a native speech-to-speech model.

When you create an agent, only name is truly required. All other fields get production-ready defaults (Groq Llama-4-Scout brain, Cartesia sonic-3 voice, Deepgram nova-3 transcriber).

Call Flows

Inbound

Code
Caller dials number → SIP provider routes to platform → platform resolves
assigned agent → agent loads config → first message plays → conversation
loop (caller speaks → STT → LLM → TTS → playback) → call ends →
call record written → webhooks fired

Key details:

• The phone number determines which agent handles the call
• First message is configurable (text, delay, or disabled)
• Voice activity detection handles interruptions and turn-taking automatically
• If transfer rules match during the call, the platform bridges to the transfer target

Outbound

Code
API call or campaign trigger → platform selects agent + caller ID →
SIP provider places call → callee answers → agent plays opening
message → conversation loop → call ends → call record written →
webhooks fired → campaign moves to next contact (if applicable)

Key details:

• Outbound calls require an agent_id and a to_number in E.164 format
• The platform selects a caller ID from your organization's assigned numbers
• Campaigns manage pacing, retry logic, and progress tracking automatically

Integration Points

Webhooks

Subscribe to events and receive POST requests at your HTTPS endpoint:

Event	When It Fires
call.started	A call has been connected and the agent is active
call.ended	A call has ended (includes duration, transcript summary)
call.ringing	An outbound call is ringing the destination
call.answered	The callee has answered an outbound call
call.failed	A call failed to connect
call.transferred	A call has been transferred to another destination
call.summary.ready	AI-generated call summary and analysis are available

Webhook payloads include a signature header for verification. See Webhooks Guide for setup.

API

The platform exposes 100+ REST endpoints grouped by resource:

Group	Endpoints	Examples
Agents	CRUD, duplicate, test call	POST /v1/agents, GET /v1/agents/{id}
Outbound	Single-call dialer	POST /v1/voice-sessions/outbound
Calls	Logs, transcripts, recordings	GET /v1/voice-sessions, GET /v1/voice-sessions/{id}
Campaigns	CRUD, start/pause/stop	POST /v1/campaigns, POST /v1/campaigns/{id}/start
Numbers	List and configure	GET /v1/numbers
Webhooks	CRUD, event listing, test	POST /v1/webhooks, GET /v1/webhooks/events
Knowledge	Files, text entries	POST /v1/knowledge-base/from-text
Billing	Balance, usage tracking	GET /v1/billing/balance, GET /v1/billing/usage
API Keys	Create, rotate, revoke	POST /v1/api-keys, GET /v1/api-keys

All endpoints require authentication via API key (Authorization: Bearer nk_live_...). See API Reference for full documentation.

MCP Tools

Agents can call external systems during a conversation using the Model Context Protocol (MCP). Configure MCP servers in the agent's tools section to let the agent:

• Look up customer records in your database
• Book appointments in a calendar
• Check order status in your OMS
• Trigger actions in any system with an MCP server

Provider Ecosystem

You can mix and match providers for each AI component:

LLM Providers

Provider	Models	Strengths
OpenAI	gpt-5.4, gpt-5.4-mini, gpt-5.4-nano, gpt-4.1, gpt-4.1-mini, gpt-4.1-nano	Highest quality, broad capability
Groq	llama-3.1-8b-instant, meta-llama/llama-4-scout-17b-16e-instruct, openai/gpt-oss-20b	Ultra-low latency inference
Phantom	phantom	Neuratel-optimized, no external API key needed

TTS Providers

Provider	Models	Strengths
ElevenLabs	eleven_flash_v2_5	Ultra-realistic voices, 100+ premium options, emotional range
Cartesia	sonic-3	Sub-100ms latency, natural prosody, speed/emotion controls
Phantom	phantom-english-speech-preview, phantom-arabic-speech-preview	Native Arabic, 10 built-in voices (aria, omar, layla, etc.)

STT Providers

Provider	Models	Strengths
Deepgram	nova-3, nova-3-medical	Industry-leading accuracy, 50+ languages, sub-200ms latency
OpenAI	gpt-4o-mini-transcribe	Strong multilingual, high accuracy
Phantom	phantom-stt-v1	Coming soon

Security

• API keys are hashed at rest — the full key is only shown once at creation
• Webhooks include a signature header for payload verification
• All traffic uses TLS encryption in transit
• Call recordings are stored in encrypted storage with configurable retention policies
• Role-based access — Admin, Operator, and Viewer roles control what team members can do and see

What's Next

• Agents — deep dive into agent configuration
• Outbound Campaigns — automated calling at scale
• Webhooks — event-driven integrations
• API Reference — full endpoint documentation
• Security — data handling, encryption, and compliance