AI Studio

Overview

The AI Studio module enables workspace administrators to build, configure, and deploy intelligent AI chatbots without writing complex backend code.

Each chatbot can leverage your workspace's database tables, document vaults, custom APIs, and serverless workflows to provide accurate, context-aware responses. AI Studio combines Large Language Models (LLMs) with Retrieval-Augmented Generation (RAG), allowing chatbots to answer questions using your organization's private knowledge while maintaining role-based security.

Whether you are building an internal support assistant, HR chatbot, sales assistant, procurement advisor, or customer-facing virtual assistant, AI Studio provides all the tools required to create, deploy, and monitor AI agents from a single interface.

Main AI Studio console displaying active agents, RAG classifications, configured models (e.g. Gemini), and management toggles.

Creating a New Chatbot

Creating a chatbot requires only a few steps.

Select AI Studio from the Site Settings menu and click New Agent.

Provide:

• Chatbot Name – A friendly name displayed to users.
• Description – A brief explanation describing the chatbot's purpose.

Once created, the chatbot is immediately available for configuration.

Creation prompt modal requesting agent names and initial descriptive roles to construct a new chatbot instance.

General Configuration

The General section defines how users experience the chatbot.

Administrators can configure:

Chatbot Profile

Configure the chatbot's identity by providing:

• Avatar Image
• Chatbot Name
• Description

These settings determine how the chatbot is presented throughout the application.

General config window showing profile avatars, system instructions prompt, custom greetings, and a live side-by-side Chat Playground.

Behavior & Personality

The System Prompt defines the chatbot's personality, behavior, response style, and operating instructions.

Typical instructions include:

• Define the chatbot's role
• Specify acceptable knowledge sources
• Control response formatting
• Restrict unsupported answers
• Define tone and language

Because the system prompt is always sent to the AI model, it becomes the primary behavior controller for the chatbot.

Welcome Message

Configure the greeting displayed whenever a user opens the chatbot.

Variables such as the logged-in user's name can be included.

Example:

`Hello {{full_name}}, how can I help you today?`

This creates a personalized user experience.

Starter Suggestion Cards

Starter cards provide users with one-click conversation starters.

Each card includes:

• Title
• Icon (Emoji)
• Theme Color
• Prompt to execute
• Short Description

When clicked, the configured prompt is automatically sent to the chatbot, allowing users to begin common conversations without typing.

Examples include:

• Procurement Dashboard
• Executive Summary
• Sales Report
• Generate Invoice
• Employee Directory

These cards improve usability by exposing the chatbot's capabilities immediately.

Suggestion card composer facilitating emoji icons, hex color themes, display descriptions, and input prompts execution templates.

Access Control

AI Studio integrates directly with CoconutDB's Role-Based Access Control (RBAC).

Administrators can specify which workspace roles are permitted to interact with each chatbot.

Examples include:

• Administrator
• Data Manager
• Editor
• Viewer
• Custom Workspace Roles

Only authorized users will be able to access and use the chatbot.

Deployment & Integration

Every chatbot automatically generates a unique Widget Embed URL.

This published URL can be embedded into:

• Internal web portals
• Customer applications
• SharePoint
• CRM systems
• ERP applications
• Corporate websites

Deployment configuration pane displaying role security checklists and widget code embed endpoints.

Widget Embed URL configuration pane detailing the generated static integration endpoint address.

User Authentication

To maintain user context and enforce workspace permissions, the embedding application must provide the user's Refresh Token when initializing the chatbot.

The refresh token enables CoconutDB to:

• Identify the current user
• Apply workspace security
• Respect role-based permissions
• Personalize responses
• Maintain conversation history
• Access user-specific data securely

Without a valid refresh token, the chatbot cannot retrieve authenticated workspace information.

Model & Credentials

The Model & Credentials section defines which Large Language Model powers the chatbot.

Administrators can configure:

• AI Model
• API Access Key

Supported models depend on the configured provider.

Examples include:

• Gemini
• OpenAI GPT
• Claude
• Azure OpenAI
• Other compatible LLM providers

The API Key is securely stored and used for all AI inference requests made by the chatbot.

Inference configuration pane allowing selection of LLM models, parameters tuning (temperature, response token lengths), and monthly limits settings.

Engine Parameters

Administrators can fine-tune model behavior by configuring:

Temperature

Controls creativity versus consistency.

• Lower values produce focused, deterministic responses.
• Higher values generate more diverse and creative responses.

Maximum Response Tokens

Defines the maximum response length that the model is allowed to generate.

Limiting response size helps control response time and AI costs.

Usage Safeguards

To control operational costs, administrators may configure usage limits such as:

• Monthly Message Limit
• Token Consumption Limits
• Request Quotas

These safeguards prevent excessive AI usage while maintaining predictable operating costs.

Knowledge Base

The Knowledge Base is the foundation of Retrieval-Augmented Generation (RAG).

Instead of relying only on the language model's pre-trained knowledge, the chatbot retrieves information from your workspace before generating a response.

AI Studio currently supports two knowledge source types.

Knowledge base dashboard listing the attached database tables and document vaults repositories connected to the agent.

Database Tables

Structured business data stored within CoconutDB tables can be connected directly to the chatbot.

For every table, administrators can configure:

• Table Selection
• Column Descriptions
• AI Usage Description

Column descriptions explain the meaning of each field, allowing the AI to better understand your data model.

Example:

`publication_date` - The official publication date of the tender.

Providing meaningful descriptions significantly improves answer accuracy.

Database table details setup modal containing field explanations and descriptive instructions detailing row properties to the chatbot.

Document Vaults

Document Vaults allow the chatbot to search uploaded files such as:

• PDF
• DOCX
• Excel
• PowerPoint
• Text Documents

Administrators simply select the vault and describe how the AI should use the documents.

Example:

`Use this document repository to answer technical specification and procurement-related questions.`

The AI will automatically retrieve relevant document content during conversations.

Document vault connection modal detailing specific operating parameters and instructions for parsing text formats.

AI Actions (Tools)

In addition to answering questions, chatbots can perform actions by invoking external APIs.

Actions extend the chatbot beyond question answering into business process automation.

Supported configurations include:

• HTTP Method
• Endpoint URL
• Headers
• Authentication
• JSON Schema
• Dynamic Parameters

Administrators define the expected input schema, allowing the AI to automatically determine when an API should be called and which parameters should be supplied.

Example actions include:

• Create Support Ticket
• Get Weather
• Create Purchase Order
• Submit Leave Request
• Generate Invoice
• Query ERP System
• Trigger CoconutDB Workflow

Because actions are schema-driven, the AI can intelligently invoke APIs during natural conversations.

API tool integration modal allowing setup of dynamic path parameters, input arguments schemas, custom headers, and request methods.

Analytics

AI Studio provides comprehensive analytics to monitor chatbot usage, performance, and operational costs.

Analytics and Audit logs console compiling conversations count, messages count, token consumptions, latency metrics, thumbs feedback, and conversation list audit history.

Administrators can filter analytics by date range and review key metrics including:

Conversation Metrics

• Total Conversations
• Active Users
• Messages Exchanged

These metrics help measure chatbot adoption.

Performance Metrics

• Average Response Latency
• Total Input Tokens
• Total Output Tokens

These statistics help optimize response speed and estimate AI costs.

User Feedback

Users can provide positive or negative feedback on chatbot responses.

Analytics displays:

• Thumbs Up
• Thumbs Down
• Overall Approval Rate

This helps continuously improve chatbot quality.

Knowledge Source Usage

AI Studio tracks how frequently each knowledge source is used.

Examples include:

• Database Tables
• Document Vaults

This helps identify which datasets contribute most to AI responses.

Chat Audit Logs

Every conversation is logged for auditing and troubleshooting.

Each record includes:

• User
• Question
• AI Response
• Feedback
• Input Tokens
• Output Tokens
• Timestamp

Audit logs can also be exported for compliance, reporting, or further analysis.

Chatbot Lifecycle

A typical chatbot implementation follows these steps:

• Create a new chatbot by providing its name and description.
• Configure the chatbot profile, system prompt, welcome message, and starter suggestion cards.
• Select the AI model and provide the required API credentials.
• Connect knowledge sources such as database tables and document vaults.
• Add external APIs or workflow actions that the chatbot can execute.
• Assign security roles to control who can access the chatbot.
• Publish the chatbot and integrate it into your application using the generated Widget Embed URL.
• Pass the authenticated user's Refresh Token during initialization so the chatbot can identify the user, enforce workspace permissions, and provide personalized responses.
• Monitor usage, performance, token consumption, and user feedback through the Analytics dashboard to continuously improve the chatbot.