Learn how Akili, Ndoto's built-in AI agent, automatically handles customer conversations using GPT-4.1, Claude, and Gemini. Configure assistants, scenarios, knowledge base FAQs, handoff rules, and custom tools.

Akili AI Agent

Akili is Ndoto's built-in AI agent. When assigned to an inbox, Akili reads incoming customer messages and responds automatically — handling common questions, routing conversations, and escalating to a human agent when needed.

Akili supports multiple AI model providers (OpenAI, Anthropic, Google, DeepSeek) and can process text, images, audio, and documents.

How Akili Processes a Message

When a customer sends a message to an inbox where Akili is active, the following happens:

Message is received — Akili picks up the incoming message, including any attachments (images, audio, documents).
Context is built — Akili collects the full conversation history to understand what has already been said.
Knowledge is searched — If FAQ lookup is enabled, Akili searches the knowledge base for relevant answers.
A response is generated — Akili sends the message to the configured AI model, which produces a reply based on the conversation context, knowledge base results, and the assistant's configured instructions.
Tools are used if needed — Akili may take actions like updating the conversation priority, adding a contact note, or routing to a specific scenario.
Response is sent — The generated reply is posted to the conversation as an outgoing message.
Handoff if needed — If Akili cannot resolve the issue or determines a human is needed, it hands the conversation off to an available agent.

Assistants and Scenarios

Akili is configured through Assistants. An assistant defines how the AI should behave for a given inbox — its personality, the product context it knows about, and any rules or restrictions.

Within an assistant, you can define Scenarios — specialized sub-agents that handle specific conversation types. For example, you might have a scenario for billing questions and another for technical support. The main assistant automatically routes conversations to the right scenario based on the context.

Scenarios can also hand back to the main assistant or escalate to a human.

Knowledge Base (FAQs)

Akili can answer questions by searching a knowledge base of FAQs. Each FAQ is a question-and-answer pair. When a customer asks something, Akili performs a semantic search across all FAQs to find the most relevant answers and uses them to inform its response.

FAQs can be:

Written manually
Auto-generated from past conversations using AI
Imported from documents (PDFs)

Akili uses vector embeddings to power semantic search — meaning it understands the meaning of a question, not just keyword matches. This is the same knowledge base that powers the Help Center search experience.

Handoff to Human

When Akili cannot confidently handle a conversation, it hands off to a human agent. This happens when:

The customer's question is outside the knowledge base
A scenario or rule triggers escalation
An AI or processing error occurs

During a handoff, the conversation status changes and the assigned human agents are notified. If configured, Akili sends the customer a message letting them know a human will be with them shortly.

Supported AI Models

Akili supports multiple AI model providers. The model used is configured per assistant:

Provider	Models
OpenAI	GPT-4.1, GPT-5
Anthropic	Claude Haiku 4.5, Claude Sonnet 4.5
Google	Gemini 3 Flash, Gemini 3 Pro
DeepSeek	DeepSeek Chat, DeepSeek Reasoner

Audio messages are transcribed using OpenAI Whisper before being passed to the model.

Configuration Options

Each assistant can be configured with:

Option	What it does
Model	Which AI model generates responses
Temperature	Controls how creative or consistent responses are (0 = deterministic, 1 = creative)
Product context	Background information about your product that Akili uses to answer questions
Response guidelines	Custom instructions for tone, style, and behavior
Guardrails	Topics or actions Akili should never engage with
FAQ lookup	Whether Akili searches the knowledge base when responding
Memory	Whether Akili remembers context across multiple conversations with the same contact

Tools

Akili can take actions during a conversation using tools. Built-in tools include:

FAQ lookup — search the knowledge base for relevant answers
Handoff to human — escalate the conversation to an available agent
Add contact note — record a finding on the customer's profile
Update priority — change the conversation priority level
Update status — resolve or reopen a conversation

Custom tools can also be built to connect Akili to external systems — such as checking an order status, looking up account details, or triggering a workflow in another service.

Multimodal Support

Akili handles more than text:

Images — included in the message context sent to the AI model
Audio — transcribed to text before processing
Documents — PDFs can be uploaded to the knowledge base and indexed for FAQ search

Frequently Asked Questions

What is Akili? Akili is Ndoto's built-in AI agent that automatically reads and responds to customer messages in your inbox, using AI models from OpenAI, Anthropic, Google, or DeepSeek.

Can Akili hand off to a human agent? Yes. Akili automatically escalates conversations to a human agent when it cannot confidently resolve the issue, when a scenario triggers escalation, or when an error occurs.

What AI models does Akili support? Akili supports GPT-4.1, GPT-5 (OpenAI), Claude Haiku 4.5, Claude Sonnet 4.5 (Anthropic), Gemini 3 Flash, Gemini 3 Pro (Google), and DeepSeek Chat and Reasoner.

Can I give Akili custom instructions? Yes. Each assistant can be configured with product context, response guidelines, guardrails, and custom tools — giving you full control over how Akili behaves.

Akili AI Agent — Automated Customer Support in Ndoto