Global Settings

Configure agent-level defaults for voice, LLM, knowledge base, and call behavior

Global settings define the default behavior for your entire flow agent. These defaults apply across all nodes unless overridden at the node level.

Overview

Global settings are organized into these categories:

Category	Purpose
System Prompt	Core AI instructions
Voice Settings	Default voice for all nodes
LLM Settings	Default language model
Noise Cancellation	Audio processing
Knowledge Base	RAG data sources
MCP Tools	Globally available tools
Outcome	Call result tracking
Phone Number	Assigned phone numbers
Call Settings	Timing and behavior defaults
Webhook	External event notifications

---

System Prompt

The system prompt defines the agent’s identity, behavior, and constraints. It applies across all conversation nodes unless a node has its own prompt that overrides it.

System prompt is required. Your flow cannot be saved without it.

The default value is You are a helpful assistant that will answer users questions.

The system prompt supports system variables and custom variables only. Extracted variables (collected during conversation) are not available here.

Enhanced Turn Taking

Toggle to enable better numeral capturing and backchannel detection for more natural conversations.

Default: Off

Prompt Enhancer

Controls automatic prompt enhancements applied to your system prompt.

Option	Description
Disabled	No enhancements — your prompt is used as-is
Basic (default)	Stable, model-agnostic prompt enhancements
Advanced	Model-aware prompts with live flow context

Example

You are a customer service representative for Acme Corp.
Today is {{current_date}} and the time is {{current_time}}.
The caller's number is {{user_number}}.
Working hours: {{working_hours}}

→ See Writing Effective Prompts for full guidance on structuring prompts and using variables.

---

Voice Settings

Select a voice from Hamsa’s voice library. The selected voice is the default for all nodes; individual nodes can override it.

Voice configuration works the same way as in single-prompt agents.

Expressiveness

Controls the emotional range and variation of the voice.

Range: 0.0–2.0 Default: 1.0

Voice Dictionaries

Attach pronunciation dictionaries for specialized terms (brand names, technical jargon, foreign words). Dictionaries are managed in the Voices section and selected here.

STT Model

Select the speech-to-text model for transcription.

Model	Description
Hamsa-STT-S2 (default)	Stable Arabic/English recognition
Hamsa-STT-S3-beta	Most recent model (beta)
Hamsa-STT-English	English-focused recognition

→ See Voice Settings for full details on browsing, filtering, previewing, and selecting voices.

---

LLM Settings

Configure the default language model for your flow. Individual nodes can override the model and temperature.

Provider and Model

Provider	Models
OpenAI (default)	GPT-5, GPT-5-Mini, GPT-5-Nano, GPT-4.1, GPT-4.1-Mini, GPT-4.1-Nano, GPT-4o, GPT-4o-mini
Gemini	Gemini 2.5-Pro, Gemini 2.5-Flash, Gemini 3.0 Flash Preview, Gemini 3.1 Flash Lite
Groq	GPT-120-OSS, GPT-20-OSS
DeepMyst	gpt-4.1-optimize, gpt-4.1-mini-optimize
Custom	Any OpenAI-compatible endpoint (requires base URL and API key)

Default: Gemini 3.1 Flash Lite

Temperature

Controls response variability.

Range: 0.0–1.0 Default: 0.2

GPT-5 family models (GPT-5, GPT-5-Mini, GPT-5-Nano) require temperature = 1.0. This is enforced automatically when you select a GPT-5 model.

Node-Level Overrides

Model and temperature can be overridden per conversation node. This lets you use a more capable model for complex reasoning nodes and a lighter model for simple confirmations.

---

Noise Cancellation

Removes background noise from the caller’s audio.

Model:

• Disabled — No noise cancellation
• Telephony Optimized — Optimized for phone call audio
• General Use Cases — Broader noise cancellation

Granularity (when enabled):

• Per Conversation — Apply once for the entire call
• Per Turn — Apply to each speaking turn separately

Additional options:

• Auto Gain Control — Normalize audio volume levels
• Send Denoised to STT — Send the cleaned audio to the speech-to-text engine

Do not enable both Noise Cancellation and Background Noise — they conflict and create audio artifacts.

---

Knowledge Base

Attach knowledge sources that your agent can query during the conversation using Retrieval-Augmented Generation (RAG).

When a user asks a question, relevant content is retrieved from the knowledge base and provided to the LLM as context.

Knowledge base items are created and managed in the Knowledge Base section and then attached to your agent here.

→ See Knowledge Base for full details on creating and managing knowledge items.

---

MCP Tools

Model Context Protocol (MCP) tools attached here are available to all conversation nodes in the flow — any node can invoke them without per-node configuration.

Use MCP tools for functions that multiple nodes need (for example, a customer lookup or a business hours check). For actions that only one node needs, use a node-level tool instead.

→ See Tools for full details on creating and configuring tools.

---

Outcome

Define how call results are classified for analytics and reporting.

Configuration

outcomeField: string          // Name of the outcome category
outcomeResponseShape?: object // Expected response structure (optional)

Example

{
  "outcome": "APPOINTMENT_BOOKED",
  "details": {
    "appointment_date": "2024-11-20",
    "appointment_time": "14:00"
  }
}

Outcomes are typically set in End Call nodes, but can also be set via tool calls or webhooks during the conversation.

---

Phone Number

Assign phone numbers to this agent for inbound and outbound calls.

Inbound: When a customer calls an assigned number, the call routes to this flow agent.

Outbound: When initiating a call via API, the agent uses one of the assigned numbers as caller ID.

All assigned numbers use the same flow configuration. To use different flows for different numbers, create separate agents.

---

Call Settings

Configure timing and behavior defaults for all calls.

Setting	Range	Default	Description
Wait For User To Speak First	Never / Always / Outbound Calls	Never	Whether the agent waits for the user to speak before responding
Agentic RAG	On / Off	Off	Agent decides when to search the knowledge base. Adds latency but improves accuracy
User Gender Detection	On / Off	Off	Detects user gender for appropriate speech form (useful for Arabic)
Smart Call End	On / Off	Off	Ends call gracefully when user asks to wrap up. Has an optional custom prompt
Language/Dialect Switcher	On / Off	Off	Switches language or dialect instantly when the user requests
Speaker Identification (Beta)	On / Off	Off	Identify and distinguish different speakers on the call
Interrupt	On / Off	Off	Whether the user can interrupt the agent while speaking
Ambient Sound	On / Off	Off	Enable ambient background sound
Thinking Voice	On / Off	Off	Play thinking/processing sound between responses
Response Delay	100–1500ms	400ms	Delay before agent responds
User Inactivity Timeout	5–60s	15s	Time before user is considered inactive
Max Call Duration	30s – 1 hour	5 min	Maximum allowed call length
Minimum Interruption Duration	0.2–1.5s	0.5s	Minimum speech length to count as an interruption
VAD Activation Threshold	0.2–0.9	0.5	Voice activity detection sensitivity

→ See Call Behavior Settings for full details on each setting.

---

Webhook

Send call events to an external HTTP endpoint.

Configuration

webhookUrl: string | null
webhookAuth: {
  authKey: 'noAuth' | 'bearer'
  authSecret?: string
}

The webhook endpoint must respond within 5 seconds. Return 200 OK quickly and process the payload asynchronously.

→ See Webhooks for event types, payload structure, and authentication details.

---

Next Steps

Node Types

Learn about all available node types

Transitions

Control how conversations move between nodes

Variables

Pass data between nodes and into prompts

Best Practices

Tips for building reliable flow agents