AI Agents

Configure artificial intelligence agents to automate customer service in Interflow.

Overview

Interflow's AI Agents are intelligent virtual assistants that can handle customers autonomously, answer questions, schedule appointments, transfer to teams, and much more.

🔄 Automatic Flow

When you create an AI Agent, the system automatically generates a linked flow. In automatic attendance, the agent runs in that flow. You can also run the same agent on a message, without starting a flow.

Available Models

In the AI Agent you choose the integration and the model:

Integration	Who pays	Models
Interflow	Plan credits	OpenAI (list below)
OpenAI (your key)	Your OpenAI account	Same OpenAI family
DeepSeek (your key)	Your DeepSeek account	DeepSeek V4 Flash and V4 Pro

OpenAI — With Reasoning

Models with advanced reasoning (effort level and verbosity configurable).

Model	Description	Cost
GPT-5.5	Premium reasoning	💲💲💲💲
GPT-5.4	Latest reasoning	💲💲💲
GPT-5.4 Mini	Economical reasoning	💲
GPT-5.4 Nano	Super economical reasoning	💲
GPT-5.2	Advanced reasoning	💲💲💲
GPT-5.1	Advanced reasoning	💲💲💲
GPT-5	Reasoning	💲💲💲
GPT-5 Mini	Economical reasoning	💲
GPT-5 Nano	Super economical reasoning	💲

OpenAI — Conversation (Chat)

Specialized in conversation; gpt-5.X-chat-latest models also support reasoning (except the legacy GPT-5 Chat Latest).

Model	Description	Cost
GPT-5.3 Chat Latest	Conversation	💲💲💲
GPT-5.2 Chat Latest	Conversation	💲💲💲
GPT-5.1 Chat Latest	Conversation	💲💲💲
GPT-5 Chat Latest	Conversation (legacy)	💲💲💲

OpenAI — Without Reasoning

Simpler and faster models, with temperature and max tokens.

Model	Description	Cost
GPT-4.1	Intelligent and versatile	💲💲💲
GPT-4.1 Mini	Low cost with good performance	💲
GPT-4.1 Nano	Most economical, light tasks	💲

DeepSeek (your own key)

Available after you register the DeepSeek integration. Does not consume Interflow credits.

Model	Description	Cost (reference)
DeepSeek V4 Flash	Fast and economical	💲
DeepSeek V4 Pro	More capable, complex tasks	💲💲

Both support Thinking Mode (chain-of-thought) with High or Maximum effort. With Thinking Mode on, temperature is ignored by the DeepSeek API.

💡 Tip

For most support cases with Interflow credits, GPT-5 Nano or GPT-5 Mini (or GPT-5.4 Nano / Mini) offer a good cost-benefit. With your own key, see also OpenAI and DeepSeek.

🚀 Coming Soon

Other providers may be added in the future (e.g. Claude, Gemini).

Accessing AI Agents

In the main menu, click AI Agents (🤖)
You'll see a list of all configured agents
Use the search to find agents by name
Click the A → Z button to sort alphabetically

Creating an AI Agent

When creating a new agent, you'll have two options:

Option 1: Generate with AI

Use our AI-assisted generation tool to create the agent's context automatically. Just describe what you want and the AI will create the complete prompt.

Option 2: Create Manually

Create the agent from scratch, configuring each detail manually.

Editor Tabs

The AI Agent editor has 6 main tabs:

1. Context

This is the main tab where you define the agent's prompt/instructions.

Available resources:

Text Editor: Write the agent's instructions, personality, and rules
Conditions in the text: branch button for if / else blocks by channel, funnel, and tags — Channel and conditions
Generate with AI: Creates context automatically based on a description
Improve: Improves existing text using AI
Out of context: customer questions that were not found in the agent context (edit mode only)
Messages and customer data: history limit and which profile fields go to the model, with optional masking
Skills: additional context concatenated into the prompt (lists, FAQs, rules), with a Show when filter
Token Usage: Track token consumption in real time (for Interflow)

2. Test

The testing tab lets you chat directly with the AI Agent before going live.

How to Use:

After adding context, click the "Test" tab
Type messages as if you were a customer
Observe the agent's responses
Check if the behavior is appropriate

Channel and conditions

The same agent can serve several channels. On the Test tab, open Simulate context to pick channel, stage, and tags. Details: Channel and conditions.

3. Files

Add media the agent can send during conversations. Organize files into folders (categories) to make them easier to find and give better context to the AI.

Supported file types:

🖼️ Images: JPG, PNG, GIF, WebP
🎵 Audio: MP3, WAV, OGG
🎬 Videos: MP4, WebM
📄 Documents: PDF, DOCX, XLSX

Folders:

Create folders to group documents (e.g. Products, Policies, Photos)
Rename a folder with the pencil icon in the header — the change applies immediately for the whole organization
Files in the folder remain accessible after renaming

4. Tools

Configure the actions the agent can run during the conversation. There are two types:

Ready-made actions — customer profile, contact, address, query API, research websites, transfer, schedule, funnel, follow-up, and more. On each one, Available when restricts the tool to channel, stage, and/or tags.
Custom actions — tools you create (parameters + what to run when triggered)

📖 See more

Full list, description of each action, and setup steps: AI Tools. Conditions by channel: Channel and conditions.

5. Knowledge Base

Add internal company documents so the AI consults them during conversations, providing accurate answers without relying on generic reasoning.

How it works:

Each registered document automatically becomes a tool available to the AI. The search is adaptive:

Short documents (up to 3 blocks) → full content returned directly
Long documents (more than 3 blocks) → RAG semantic search returns only the most relevant excerpts

The search uses hybrid technology combining vector embeddings with keyword search, supporting Portuguese, English, and Spanish.

How to add a document:

In the Knowledge Base tab, click "Add document"
Fill in:
- Title – Document name (e.g., Business Hours)
- Description – Content summary; the more specific, the better the AI knows when to call it (e.g., Contains operating hours, holidays and phone for scheduling)
- Language – Main content language
- Content – Full text, with Markdown support
Click "Save"

💡 Tip

The description is the most important field. Use words the customer would use when asking: instead of "General information", write "Contains address, hours, accepted plans and consultation fees".

📂 Organization by subject

Prefer one document per subject (e.g., Hours, Address, Cancellation Policy) instead of one big document with everything. Focused documents generate more accurate answers.

6. Advanced

Detailed agent settings:

Basic Information

Title: Agent identifier name
Default Prompt: Make available to all organizations (superadmins only)

AI Settings

Integration: Interflow (plan credits), OpenAI (your key), or DeepSeek (your key)
Model: Select the model for the chosen integration
Temperature (OpenAI without reasoning): Controls response creativity (0.0 to 2.0)
Reasoning Level (OpenAI with reasoning): Low, Medium, or High
Thinking Mode (DeepSeek): Enable/disable chain-of-thought and effort (High / Maximum)
Max Tokens (OpenAI without reasoning): Response token limit
Verbosity (GPT-5 models): Controls response length
Timezone: Sets the timezone for dates and times

Message sending

On the Sending tab:

Send messages from the agent: the final reply goes straight to the chat (remove the Send Text node from the flow to avoid duplicates)
Split text by paragraph: each paragraph (blank line) becomes a separate message
Signature: text at the beginning or end of the message; the model does not read this text
Reply with audio: turns the reply into audio with a system voice; on failure, sends the text
Send audio through the auxiliary channel: on WhatsApp Official with a connected WAHA auxiliary, audio goes out through the auxiliary (OGG, more natural sound)

Flow Settings

View Flow: Access the agent's linked flow
Reset Flow: Recreates the flow with the default configuration
Triggers: Configure when the agent should be activated

Fallback Integration

Configure a secondary AI agent that automatically takes over when the primary agent cannot respond or reaches a limit.

What it's for

Ensure 24/7 service even when the primary agent is limited
Automatically redirect to a specialized agent for certain topics
Prevent customers from going without a response in error or timeout cases

How to configure

In the agent's Advanced tab, locate the "Fallback Integration" section
Select the fallback AI Agent
Configure desired alerts (notification when fallback is triggered)
Define use cases for automatic activation
Save

AI Credits

AI Agent usage is measured in USD-based credits, providing transparency in cost control.

Credit Types

Type	Description
Plan Credits	Deposited on monthly renewal. Reset when renewed.
Purchased Credits	Acquired separately. Do not expire.
Bonus Credits	Promotional or referral. Do not expire.

Cost Table by Model (Interflow credits / OpenAI)

Model	Input (1M tokens)	Output (1M tokens)
GPT-5.5	$5.00	$30.00
GPT-5.4	$2.50	$15.00
GPT-5.4 Mini	$0.75	$4.50
GPT-5.4 Nano	$0.20	$1.25
GPT-5.2 / GPT-5.3 Chat Latest	$1.75	$14.00
GPT-5.1 / GPT-5 / GPT-5 Chat Latest	$1.25	$10.00
GPT-5 Mini	$0.25	$2.00
GPT-5 Nano	$0.05	$0.40
GPT-4.1	$2.00	$8.00
GPT-4.1 Mini	$0.40	$1.60
GPT-4.1 Nano	$0.10	$0.40

DeepSeek pricing (billed on your DeepSeek account): see DeepSeek Integration.

💡 Savings

Cached tokens cost up to 10x less. Mini and Nano models are significantly cheaper for high-volume use cases.

Credit Statement

Access the detailed history at Settings → Usage → AI Credits:

Total available balance
Breakdown by source (Plan, Purchased, Bonus)
Full transaction history
Filters by transaction type and usage

Signature in Messages

Add an automatic signature to messages the agent sends, in the same format as the Send Text node. It appears in the chat for the customer, but it is not included in the history the model reads — the AI will not copy or duplicate the text.

How to configure

Open the AI Agent on the Sending tab
Turn on Send messages from the agent
Fill in Signature and choose Beginning (Header) or End (Footer)
If the flow still has a Send Text node after the agent, remove it to avoid duplicate messages

Include line breaks in the signature text if you want it separated from the content.

Older flows

With Send messages from the agent off, sending still happens in the Send Text node — the signature in that case is still the node's. See Text Message.

Changelog: v2026.8.9

Audio replies

The AI Agent can send the final reply as audio, without TTS nodes in the flow. Register the voice in System voices (ElevenLabs, Minimax, or OpenAI) and turn the option on in the Sending tab.

How to configure

Open the AI Agent on the Sending tab
Turn on Send messages from the agent
Turn on Reply with audio and choose the voice
Adjust the rules:
- If the customer sent audio, reply with audio
- If the reply contains a link, send text
- Audio chance on text replies (0–100%; 25% is 1 in 4)
- Send audio through the auxiliary channel — on WhatsApp Official with a connected WAHA auxiliary, audio goes out through the auxiliary (OGG conversion and a more natural sound). If the auxiliary is unavailable, it sends through Official.
If the flow still has TTS / audio nodes after the agent, remove them to avoid duplicate sends

The model text is rewritten for speech (numbers, dates, and URLs spoken out). The customer receives the audio with that version as the transcription. Interactive lists and media JSON stay as text.

If audio fails

Any error (voice, rewrite, TTS, upload, or send) makes the agent send the original text. The customer still gets a reply.

Changelog: v2026.9.5

Intelligent Follow-up

The AI can automatically adjust follow-ups based on the conversation context.

How It Works

When the customer mentions a future date or indicates this isn't the right time:

"Let's talk Thursday" → AI reschedules follow-up to Thursday
"I'll be back on the 25th" → AI schedules return for the 25th
"I've already closed a deal" → AI pauses the follow-up

Configuring

Access the agent's Tools
Add the "Change Return Date" action
Select the Flow and Node destination
Save

Transfer with Automatic Summary

When transferring a conversation to a human team, the AI can automatically generate a summary:

Edit the "Transfer to Team" action
Check: ✅ "Generate attendance summary on transfer"
Save

The summary includes:

Reason for contact
Main customer questions
Steps already taken
Important data collected

Signature in Messages

Add an automatic signature to AI-generated messages.

How to Configure

The signature is configured in the linked flow, specifically in the Send Text Node.

Access the AI Agent's linked flow
Locate the Send Text node that sends AI responses
In the message field, add the desired signature after the response variable

Example:

text

{{ai_response}}

---
🤖 Message sent by AI

Scheduling Integration

The AI can manage appointments conversationally:

Example Conversation

text

👤 Customer: I'd like to schedule an appointment
🤖 AI: Sure! What day works for you? We have slots Monday through Friday.

👤 Customer: Thursday at 2pm
🤖 AI: We have availability Thursday at 2pm with Dr. John. 
       Shall I confirm the appointment?

👤 Customer: Yes, please
🤖 AI: ✅ Appointment confirmed!
       📅 Date: Thursday, 12/26/2025
       ⏰ Time: 2:00 PM
       👨‍⚕️ Professional: Dr. John
       
       I'll send a reminder 24h before. See you then! 😊

Messages and customer data

Each agent controls how much history and which customer data go into the context sent to the model. On the Context tab, Messages and customer data opens these settings.

The message limit is how many messages from the current chat the model sees (1 to 200; default 50). On long conversations, a lower limit cuts tokens without changing the prompt.

Under Customer data, choose what goes in the first message after the prompt:

Option	What the model receives
Yes	All profile fields. Optional: turn on Mask data and pick what to partially hide (tax ID, email, phone, address)
No	No profile data
Some	Only the fields marked Send; on each one, you can also mark Mask

Available fields: name, email, document (tax ID), country, contacts, addresses, tags, funnel stage, custom fields, previous conversations, and upcoming appointments. Channel and system date/time are always included, regardless of this option.

Existing agents without this setting keep 50 messages and all data, unmasked.

Changelog: v2026.8.21

Run on a message

In the menu of a customer or agent message, Run AI agent fires the prompt without starting a flow. The model sees only that message (plus date, channel, and customer data). Tools are the same; sending to the chat depends on Send messages from the agent.

It does not appear on a private note, a system event, or a scheduled chat. Permission matches starting a flow.

Full guide: Run on a message

Skills

Skills are additional context concatenated into the agent's prompt — product lists, FAQs, rules, or any extra block. Unlike the knowledge base, the content goes straight into context; the AI does not need to call a tool.

Types

Free text: FAQ, rules, catalog
WhatsApp List: Formats content for interactive WhatsApp lists

How to add

On the Context tab, under Skills, click Add skill
Choose the type (free text or WhatsApp list)
Fill in title, description, and content
(Optional) set Show when — channel, stage, and/or tags
Save the agent context

Matching skills are concatenated into a single text. With no filter, the block is always included. Guide: Channel and conditions.

Best Practices

✅ Do

Be specific in the prompt about expected behavior
Provide examples of ideal responses
Limit the scope of action (what AI can and cannot do)
Test extensively before going live
Monitor conversations and adjust the prompt
Use Out of context to review customer questions the agent did not find in the context

❌ Avoid

Vague or generic prompts
Allowing the AI to invent information
Excessive autonomy without supervision
Forgetting to configure transfer to human
Ignoring customer feedback about AI

Metrics and Reports

Track your AI Agents' performance:

Resolution rate: % of conversations resolved without human
Average response time: AI speed
Transfers: How many times it transferred to a human
Satisfaction: Customer ratings
Token usage: API consumption

Next Steps

Run on a message - Run the agent without a flow
Channel and conditions - One agent on several channels
AI Tools - Configure agent actions
Text Enhancer - AI commands for agents
Flows - Integrate AI into flows
Schedule - Configure AI-powered appointments
Settings - AI controls

AI Agents ​

Overview ​

Available Models ​

OpenAI — With Reasoning ​

OpenAI — Conversation (Chat) ​

OpenAI — Without Reasoning ​

DeepSeek (your own key) ​

Accessing AI Agents ​

Creating an AI Agent ​

Option 1: Generate with AI ​

Option 2: Create Manually ​

Editor Tabs ​

1. Context ​

2. Test ​

3. Files ​

4. Tools ​

5. Knowledge Base ​

6. Advanced ​

Basic Information ​

AI Settings ​

Message sending ​

Flow Settings ​

Fallback Integration ​

What it's for ​

How to configure ​

AI Credits ​

Credit Types ​

Cost Table by Model (Interflow credits / OpenAI) ​

Credit Statement ​

Signature in Messages ​

How to configure ​

Audio replies ​

How to configure ​

Intelligent Follow-up ​

How It Works ​

Configuring ​

Transfer with Automatic Summary ​

Signature in Messages ​

How to Configure ​

Scheduling Integration ​

Example Conversation ​

Messages and customer data ​

Run on a message ​

Skills ​

Types ​

How to add ​

Best Practices ​

✅ Do ​

❌ Avoid ​

Metrics and Reports ​

Next Steps ​

AI Agents

Overview

Available Models

OpenAI — With Reasoning

OpenAI — Conversation (Chat)

OpenAI — Without Reasoning

DeepSeek (your own key)

Accessing AI Agents

Creating an AI Agent

Option 1: Generate with AI

Option 2: Create Manually

Editor Tabs

1. Context

2. Test

3. Files

4. Tools

5. Knowledge Base

6. Advanced

Basic Information

AI Settings

Message sending

Flow Settings

Fallback Integration

What it's for

How to configure

AI Credits

Credit Types

Cost Table by Model (Interflow credits / OpenAI)

Credit Statement

Signature in Messages

How to configure

Audio replies

How to configure

Intelligent Follow-up

How It Works

Configuring

Transfer with Automatic Summary

Signature in Messages

How to Configure

Scheduling Integration

Example Conversation

Messages and customer data

Run on a message

Skills

Types

How to add

Best Practices

✅ Do

❌ Avoid

Metrics and Reports

Next Steps