AloAi Agents are AI-driven virtual assistants designed to handle customer interactions over voice calls and text messaging. They help automate common workflows such as answering frequently asked questions (FAQs), sending follow-ups, and escalating complex issues to live agents when needed.
By handling repetitive tasks, AloAi Agents let businesses keep communication consistent while reducing the manual work agents would otherwise do.
AloAi agent types: Voice and text
AloAi Agents come in two types, each built for a different interaction channel.
Voice agents handle phone-based workflows. They make outbound calls, answer inbound calls, route calls as needed, and perform follow-ups.
Text agents handle SMS communication. They send messages, respond to incoming text replies, deliver reminders, and update customers over SMS.
To see how AloAi agents can enhance your business operations, check out this blog.
Getting started with AloAi Agents
Admins can create and configure AloAi Agents directly from their Aloware admin accounts.
Setup involves selecting the agent type, defining the agent's behavior, assigning communication lines, and applying automation settings.
To create a new agent:
Log in to your Aloware admin account.
Open the AloAi Agents menu, where you manage existing agents or create new ones.
Click the +New Agent button to start configuring a new bot.
Choose Voice agent or Text agent, based on the workflow you want to automate.
Specify whether the agent handles inbound tasks (answering incoming calls or texts) or outbound tasks (sending follow-up messages).
You can also view AloAi Agents from the Users menu, which lets admins see agent assignments and related user settings in one place.
If this is your first agent, use Aloware's interactive walkthrough for in-platform setup guidance.
Naming the agent
Give the agent a clear name and describe its purpose. A specific name and description make agents easier to tell apart when you manage several at once.
For example, name a customer support agent "Support bot."
Selecting the AI model
Choose an AI model based on how complex your call handling needs to be. The three providers differ in their strengths.
Claude (Anthropic) suits scenarios that need strong contextual understanding, such as legal, healthcare, or other sensitive use cases.
OpenAI works well for fast, conversational call flows and interactive voice experiences.
Gemini (Google) is built for lightweight, speed-focused interactions.
After picking a provider, select the specific model and tier the agent uses when handling calls and texts.
Choose a faster model when response speed matters, or a more capable model when the agent needs to handle in-depth conversations. Available models include:
Field Label | AI model for calls + Tier | Text model |
Anthropic |
| - |
| - | |
OpenAI |
|
|
OpenAI Realtime |
| - |
Setting voice, language, and tone
Voice profiles let you select a tone, such as professional, casual, or neutral, which shapes how the agent sounds on calls.
Language and voice settings let the agent interact with customers across regions.
Supported languages include:
English (US/UK)
Spanish (LATAM/ES)
Multilingual (English and Spanish)
Spanish (LATAM and ES)
French
Chinese
German
Japanese
Portuguese (Portugal and Brazil)
Russian
Italian
Korean
Dutch
Polish
Turkish
Vietnamese
Romanian
Writing instructions
Instructions act as the agent's internal guide. They shape its tone, how it interprets customer input, and how it responds across different scenarios.
Clear instructions keep the agent consistent, help it follow your business logic, and let it handle both simple and complex interactions. Instructions also control how conversations begin, especially on outbound calls.
When writing custom instructions, include:
The tone the agent should use.
Keywords or phrases to prioritize or avoid.
Routing or escalation logic.
Guidance for questions the agent cannot answer.
Read this guide to learn more how to write instructions.
Speak First
Speak First makes the voice agent deliver its greeting as soon as an outbound call connects, rather than waiting for the contact to speak. This removes the initial silence at the start of the call.
You can add a delay of 1 to 60 seconds so the greeting does not overlap with the contact's first words. Speak First applies only to outbound calls. Inbound calls follow their default flow.
Greeting and goodbye message
The greeting is the agent's opening message to the customer. You can include variables such as the customer's first name, for example: "Hi [First Name], how can I assist you today?"
Choose one of two greeting types:
Static (custom message) plays the exact message you write every time. Use it when you want the opening to be consistent and predictable.
Dynamic (AI-generated) has the AI generate the greeting based on context, such as the contact's details or conversation history. Use it when you want the opening to adapt to each customer.
The goodbye message is the closing line the agent uses when ending a conversation.
Adding functions to AloAi Agent
Functions are automation tools the agent runs during a conversation. The agent triggers them based on customer replies or detected intents, which lets it complete tasks without human involvement. Functions are grouped by category.
Call control
Leave voicemail (for voice agents) sends the call to voicemail and ends the AI flow.
Transfer call (for voice agents) routes the call to a live agent or another team.
Route to extension (for voice agents) lets callers reach a team member by speaking an extension number.
End call ends the interaction.
Contact management
Add contact disposition labels the contact based on the interaction outcome.
Add contact tags applies tags for segmentation and reporting.
Add to list adds contacts to lists for campaigns or tracking.
Update contact field updates customer details in the CRM.
Disengage contact ends the conversation automatically when needed.
Set contact as DNC (Do Not Contact) (for voice agents) marks contacts who request no further outreach.
Communication
Send SMS (for voice agents) sends a follow-up or confirmation message.
Add communication tag tags the interaction type for reference and analytics.
Call management
Summarize as note creates a summary of the interaction and saves it to the contact record.
Add call disposition records the outcome of a voice interaction.
Scheduling
Manage appointment schedules or updates an appointment using predefined details.
Enroll to sequence adds contacts to an automated sequence or workflow.
Execute SMS sequence (for text agents) runs an SMS sequence for the contact.
Integrations
Create HubSpot ticket opens a ticket in HubSpot.
Route to HubSpot deal contact owner directs calls to the assigned CRM owner. This requires a pipeline selection and a fallback inbox for cases where no owner is available.
Advanced
Custom functions trigger API-based actions tailored to your workflows.
Understanding context across conversations
AloAi Agents use shared contact context to maintain continuity across interactions, even when conversations happen on different channels or with different agents.
Each interaction is tied to the contact, not to a specific session or bot.
With this setup, agents can access:
Contact information such as name, email, phone, and address.
Communication history including previous calls, messages, insights, and interactions.
Appointment information for upcoming and, optionally, past appointments, including details and availability.
HubSpot deal information such as deal stage, amount, and close date, retrieved live from HubSpot at call time for the caller's associated contact record. For a detailed explanation of how this works, see our guide.
Guesty reservation information pulled directly from Guesty in real time.
How Guesty reservation context works
When a guest asks a question such as "What's the WiFi password?" or "What's the lockbox code?", the agent retrieves the answer from Guesty in real time, so guests get answers without waiting for your team.
As long as the Guesty integration is enabled and the agent has access to the required settings, the agent can retrieve reservation details such as the lockbox code, WiFi password, property or unit name, and check-in and check-out dates. The agent applies the following rules:
If the reservation cannot be found, the agent asks the guest to verify their reservation details before continuing.
If the guest has not yet checked in, the agent confirms the reservation but withholds sensitive details, such as lockbox or access codes, until the stay becomes active.
If reservation details are missing in Guesty, the agent tells the guest the information is currently unavailable rather than providing incomplete or inaccurate details.
This reservation context is shared across both voice and text agents, so guests do not repeat themselves when they switch channels, and your team continues the conversation with the full reservation background.
Lockbox and access codes are generated per guest and require Guesty Locks Manager to be configured. For setup instructions, see Managing lock code settings in Guesty Locks Manager.
For accounts that do not use Guesty Locks Manager, the agent can instead read the lock code from a designated property custom field in Guesty. The custom field key follows your agreed naming convention or can be configured per account. For full setup steps, see Setting up Guesty reservation context for AloAi Agents.
What this means for admins
Admins control how context is shared and displayed within Aloware. Specifically, you can:
View and manage unified contact context by accessing a complete view of each contact, including messages, call logs, and appointment details.
Reassign AloAi Agents without losing data by switching bots or updating workflows without resetting contact history.
Configure data retrieval by deciding whether historical context, such as past appointments or messages, is available, based on your compliance or operational needs.
Enable or disable context sharing by channel (voice or text) or per agent.
What this means for users or agents
Shared context gives the agent the background it needs for an informed conversation. The agent can answer requests for account details, appointments, or summaries of past conversations with up-to-date information.
Customers can also switch between voice and text bots, or between different agents, without restating who they are or what was discussed before.
Connecting to MCP servers
An agent can connect to an external MCP (Model Context Protocol) server and call that server's tools during a live conversation.
The agent acts as the MCP client: you point it at a server's URL and tell it how to authenticate, and from then on the agent can call the tools that server exposes while on a call.
Whatever the external server can do, such as look something up, create a record, or trigger a workflow, becomes an action the agent can take in the moment, without that logic living inside Aloware.
You configure each connection per agent on the Add MCP form. Once saved, that server's tools are available to the agent.
The Add MCP form
The Add MCP form has the following fields:
MCP Name is a label for the connection inside the agent's configuration. Use something that identifies which server it points at.
MCP URL is the endpoint of the MCP server the agent connects to. The eye toggle masks or reveals the value, so treat it as sensitive.
Timeout (ms) is how long the agent waits on the server before giving up, in milliseconds. It defaults to 10000, or 10 seconds.
Authentication is how the agent authenticates to the server. Three methods are available:
Bearer Token (Headers) sends a bearer token as an HTTP header on each request. Add the token under Headers, for example "Authorization: Bearer <token>." This is the default.
OAuth 2.1 authenticates through an OAuth 2.1 authorization flow instead of a static token.
No Authentication connects to the server without credentials, for open or unauthenticated servers.
Headers are HTTP headers sent with the connection request, entered as key/value pairs. Add more with +New key value pair. The bearer-token header goes here when you use header authentication.
Query Parameters are query-string parameters appended to the MCP URL, entered as key/value pairs. Add more with +New key value pair.
Relay through Dashboard controls where tool calls originate. It is OFF by default.
This toggle controls where the outbound tool call originates. When it is off, the default, the agent connects to the server directly.
When it is ON, calls route through the Aloware dashboard, which keeps the connection's credentials server-side rather than using them directly from the agent runtime. Turn it on whenever the server's credentials are sensitive.
Security considerations
Giving an agent the ability to call external tools introduces the same risks as any model-plus-tools setup.
The risk most specific to agent workflows is prompt injection, where untrusted content the agent encounters, such as what a caller says, transcript text, or documents, carries instructions that try to trick the agent into calling a tool it should not.
To reduce this risk:
Relay credentials through the dashboard. Turn on Relay through Dashboard for any server with sensitive credentials, so keys stay server-side.
Use least privilege. Point the agent at a server or key scoped to the minimum it needs.
Prefer read-first tools. Favor connections that read before they write, and gate destructive actions behind explicit intent.
Be careful with PII. Tool calls may pass contact data to the external server, so send only what the task needs.
Test against non-production servers. Validate a new connection against a sandbox or dev server before pointing a live agent at it.
Troubleshooting MCP connections
Common issues and their fixes:
Tools are not available to the agent. Verify the MCP URL is reachable and the connection saved, and confirm the authentication method and any required headers.
Authentication or 401 errors. Check the bearer token, or your chosen auth method, and that the token header is entered correctly under Headers.
Calls time out. Raise the Timeout (ms) value if the server is slow to respond.
Cal.com and Calendly calendar integrations
A voice agent can book appointments through Cal.com or Calendly.
Both connect the same way: the admin generates an API key inside their own Cal.com or Calendly account and pastes it into the agent.
From that point, the provider is the calendar engine. It owns the events, the hosts, the availability rules, and the notifications, and the agent reads those events over the API and books appointments on the contact's behalf during a live call.
The agent never defines availability itself. It only books against events the admin has already created in the provider. Events come in three shapes:
Individual is a single host's appointment.
Round-robin distributes bookings across multiple hosts based on availability.
Collective or conference has multiple hosts attend the same booking together.
Plan requirements
The plan you need depends on the provider and the event types you book. Confirm current pricing and feature gating against each provider's documentation, since third-party plans change.
Calendly requires a paid plan for the agent to book at all. On the Free plan, API access is read-only, and booking through the Scheduling API plus webhooks requires a paid tier (Standard, Professional, Teams, or Enterprise).
Cal.com issues API keys on every plan, including Free, so a single-user calendar can book for free, and automated reminder workflows are included on all plans. Round-robin and collective or conference event types require the paid Teams plan.
For any multi-user or team deployment, plan on a paid subscription with both providers. The only setup that runs entirely on a free plan is a solo Cal.com calendar.
Setting up the integration
Configuration happens in two places: the provider and the agent. To connect a calendar:
Create or use a Cal.com or Calendly account on a plan that supports the event types you need.
Create the events the agent should book against (individual, round-robin, or collective or conference) across the users who take bookings. Configure all timing, location, and conferencing settings on the event itself.
Generate an API key in the provider.
In Cal.com, go to Settings, then Developer, then API keys, then Create.
In Calendly, go to Integrations, then API & Webhooks, then generate a personal access token.
Paste the API key into the agent's configuration.
Cal.com API key
Calendy API key
Once the key is saved, the agent can see every event in the connected provider account and books through that key. There is no per-event wiring to do on the Aloware side. The provider remains the system of record for the calendar.
Note - When you use the Cal.com or Calendly integration to manage appointments, you do not need to add the Manage appointment function in the instructions. The integration handles the booking.
Booking behavior
The event's configuration in the provider drives all behavior, including duration, location, conferencing, round-robin versus collective, buffers, and notice periods.
During booking, the agent calls the provider over the API, looks up the event, checks the availability of the users attached to it, and schedules into an open slot.
After booking, the invite is delivered based on your integrations:
If the agent or account is also integrated with Google Calendar, the invite is delivered to Google Calendar.
Otherwise, Cal.com or Calendly sends the invite by email to the booked contact and to the users on that event.
If the event is set up as a conference or collective, attendees are placed into that conference.
Notifications and reminders
Notification settings live in the provider, not in Aloware. Confirmation emails, calendar invites, and reminders are configured at the event level in Cal.com or Calendly through each provider's Workflows.
When the agent books through the API, the provider's standard actions fire automatically, exactly as they would for a booking made through the provider's own interface.
SMS reminders are supported and are configured through the provider's reminder or Workflow settings on the event, not in the agent. Two provider differences apply:
Calendly text reminders fall under Workflows, which are a paid-plan feature.
Cal.com automated reminder workflows are available on all plans, including Free, though SMS sending may still be subject to the provider's own SMS limits.
Knowledge base files
The Knowledge Base lets admins add and manage knowledge sources inside the configure tab so the agent can answer questions more accurately.
Admins can create a new Knowledge Base or reuse an existing one, then populate it by uploading files, creating structured FAQ or free text entries, and importing content from webpages.
Each knowledge item moves through a processing lifecycle, and the system displays its status as Pending Processing, Processing, Success, or Failed so admins can see whether the content is ready to use.
Creating or selecting a Knowledge Base
Before adding content, attach a Knowledge Base to the agent. You can create a new one or select an existing one.
To create a new Knowledge Base:
Click +Create new Knowledge base.
Enter a name for the Knowledge Base.
Select an embedding provider: AWS Bedrock (faster) or OpenAI (default).
To use an existing Knowledge Base:
Click +Select Knowledge Base.
Choose from the existing Knowledge Bases. You can reuse a Knowledge Base from another AloAi Agent, so you do not need to upload the same files again.
Confirm your selection.
Importing webpages
Importing a webpage adds that page's content to the Knowledge Base so the agent can reference it. To start, click Add Web Pages and enter the URL.
From there, choose one of two options:
Single page imports only the page at the URL you entered. Click Confirm to import it.
Auto crawl finds all pages of the website.
Select the pages to include, then click Confirm.
The URL processes and shows its status: the current processing state, a success confirmation when complete, and a clear error message if an issue occurs.
Invalid URLs are flagged immediately.
Once a webpage imports successfully, its content becomes part of the agent's knowledge.
Uploading files
Uploaded documents give the agent material to reference when answering questions.
These can include FAQs, manuals, troubleshooting guides, structured datasets, and internal documentation.
The Knowledge Base accepts PDF, DOC DOCX, TXT, CSV, XLSX, XLS, HTML, MD, PPTX, PPT, RTF, XML, and JSON files. You can upload up to 10,000 files, with a maximum size of 50 MB per file.
Adding free text or FAQ entries
Free text entries let you create structured knowledge by hand, similar to FAQ pages, intake questions, or searchable help content.
Use them for information that is not in an uploaded file but should still be available to the agent.
Each entry includes a file name and text content field and supports up to 5,000 characters.
Open the +Add Text modal and enter the file name and text content.
Click Save. The system begins processing the entry automatically.
You can create multiple entries, edit existing ones, or delete them. After processing completes, the entry becomes searchable and available to the agent. To confirm an update, ask the agent a question related to the new or revised content.
Speech settings
Speech settings control how the agent sounds and how it handles speech timing during a conversation.
Normalize text for speech converts numbers, currency, and common abbreviations to spoken form before synthesis, for example "$50" becomes "fifty dollars." This complements Voice AI best practices.
Background sound sets the background audio during calls. Options are No sound, Coffee shop, Convention hall, Summer outdoor, Mountain outdoor, Static noise (a subtle line effect that mimics call quality), and Call center.
Smart pause detection waits for the caller to finish speaking before the agent responds.
Interruption sensitivity controls how the agent reacts when interrupted. Set high sensitivity when the agent should pause as soon as the user cuts in, which suits live interactions. Use low sensitivity when the agent needs to finish its full response, such as delivering detailed instructions.
Minimum interruption words controls how sensitively the AI can be interrupted by human speech.
Enable Back-channeling adds verbal cues such as "uh-huh" or "I see" to make conversations feel more interactive.
Reminder messages control how often the AI sends a reminder message when the user is silent, for example every 10 seconds for up to 3 times.
Call settings
Call settings define how long the agent stays on a call and how it handles silence.
Voice AI best practices injects voice AI guidelines for natural speech, pronunciation, and safe conversations.
End call on silence disconnects silent or idle calls after a set duration, such as 30 seconds.
Max call duration sets a maximum time limit to avoid prolonged or non-actionable calls.
Realtime transcription settings
Realtime transcription settings control how the agent converts speech to text during a call.
Denoising mode filters out unwanted background noise. It offers two options:
Remove noise applies noise suppression. Adjust the suppression strength to control how aggressively background noise is removed. Lower values sound more natural but filter less noise.
No denoising leaves the audio unfiltered.
Transcription mode balances speed and accuracy. It offers three options:
Optimize for speed favors faster transcription.
Optimize for accuracy favors more accurate transcription.
Custom settings exposes individual controls:
Endpointing sets the milliseconds of silence before the system considers speech finished.
Interim results returns non-final transcription results for faster response.
Punctuation adds punctuation to the transcription output.
Smart formatting applies formatting to numbers, dates, and entities.
Filler words includes filler words such as "um" and "uh" in the transcription.
Boosted keywords lets you provide a custom list of keywords to expand the model's vocabulary.
Voicemail settings
Voicemail detection applies to outbound voice agents only. It lets the agent recognize when an outbound call reaches voicemail and decide what to do next.
When the agent detects voicemail, it can:
Hang up and end the call without leaving a message.
Leave a message if reaching voicemail
Static sentence by playing a pre-recorded voicemail, which can include dynamic fields such as the customer's name.
Prompt by using an AI-generated voicemail prompt that adapts based on previous conversations with the customer.
Post call data extraction
Post-call data extraction pulls structured fields out of an agent's calls after they end, such as a call summary, a customer type, whether an appointment was booked, or a number.
After a call ends, a selectable extraction model reads the conversation and fills in each field you defined.
The value a field returns depends on its type: free text, one value from a fixed set of choices, a yes or no, or a number.
Agent-specific and shared fields
Fields come in two layers, and an agent always captures both.
Agent-specific fields are added on this panel and captured on this agent's calls only.
Shared fields apply to every agent in the organization and are managed centrally under Settings, then Post-Call Data. You do not edit them from this panel.
An agent with no agent-specific fields of its own still captures the organization's shared fields.
Field types
The +Add button offers four field types. All four share the same configuration form, and Selector adds a list of choices.
Text captures free-form text, such as a detailed call summary.
Selector captures one value from a defined list of choices, such as a customer type.
Yes/No captures a yes or no value, such as whether an appointment was booked.
Number captures a numeric value, such as a user's age.
Configuring a field
Every field type opens the same form. Fill in the following:
Name is the human-readable label for the field, such as Detailed Call Summary.
Key is the machine-readable identifier in snake_case, such as call_summary. This is the form used in data, exports, and APIs. The sparkle button generates a key from the name. Keep the key consistent and stable once a field is in use.
Description is a plain-language instruction telling the model what to capture, such as "Did the user book an appointment?" The model relies on this to know what to pull, so write it as a clear question or instruction rather than just a label.
Optional controls whether the field can be left blank. When it is on and the model cannot infer the value from the conversation, the field is left unpopulated. It is off by default.
Choices (Selector only) is the set of allowed values the model must choose from. Add rows with +Add and remove them with the trash icon.
Choosing the extraction model
The model selector picks which model performs the extraction. It defaults to GPT-4.1-mini and is searchable. Available models span two families:
OpenAI GPT | GPT-4.1 GPT-4.1-mini (default) GPT-4.1-nano GPT-4o GPT-4o-mini GPT-5.5, gpt-5.4 GPT-5.4-mini GPT-5.4-nano GPT-5.2 GPT-5.1 GPT-5 GPT-mini GPT-5-nano |
Anthropic Claude | Claude-Opus-4-8 Claude-Sonnet-4-6 Claude-Haiku-4-5 |
This list changes as models are added or retired, so treat the live picker as the authoritative source.
Security and fallback settings
Safety guardrails are an optional content-moderation layer you can turn on for an AloAi Voice Agent.
When enabled, they classify and intercept risky content in real time on both sides of the conversation: what the caller says and what the agent is about to say.
During each call, the system evaluates the conversation before the agent generates a response or triggers an action, so flagged content is caught before it reaches the caller.
Guardrails are off until you set them up, and you choose which categories to apply, so you can enable only the ones that fit your use case. For full details, see AloAi Voice Agent safety protections.
Dynamic variables
Dynamic variables let the agent pull live values into its instructions and messages instead of using fixed text. There are three types.
Contact fields pull information about the contact stored in Aloware, such as name, last name, email, and city. The agent uses whatever is filled in on the contact record in Aloware.
System variables pull specific data the agent needs during the call rather than contact details, such as current time, current hour, and current calendar.
Custom variables are values you define under Dynamic Variables setup for your own use cases. Each one has a name and a default value. For example, a "Say sorry" variable can hold a default apology sentence the agent reuses.
Test your agent
Before going live, test the agent the same way a customer would reach it. There are two ways to test, matching the agent's channel.
Call test lets you talk to a voice agent. Click the Call icon and start talking to your agent.
Text test lets you chat with a text agent. Send a message to start chatting with your agent.
