An AI assistant and an AI agent don’t play the same role in your organization.

What Is an AI Assistant?
An assistant is mainly focused on conversation and information access:
- Answers questions in natural language
- Searches for information in your data (FAQ, knowledge base, CRM, documentation, etc.)
- Helps write emails, summaries, tickets, reports, and internal documentation
- Does not usually take complex autonomous actions in your tools

In short, an AI assistant is centered on information retrieval and content generation.

What Is an AI Agent?
An AI agent goes further: it reasons, makes decisions, and takes actions in your systems.
- Connects to your tools (Salesforce, Dynamics,Zendesk, GitHub, Jira, Notion, etc.)
- Follows predefined workflows (create/update tickets, answer customers, escalate to humans…)
- Chains several steps: read data → analyze → decide → execute
- Can run in human-in-the-loop mode (agent proposes, human validates) or in full automation

An AI agent doesn’t just answer questions; it executes business tasks based on your data and internal rules.

The time it takes to connect your data depends mainly on your data sources and existing access rights.

1. Fastest Connections (Minutes to a Few Hours)
Some integrations are almost “plug-and-play”:
- Slack / Microsoft Teams – OAuth connection, choose workspaces or channels
- Google Drive / SharePoint / OneDrive – connect drives, folders, and files
- Notion / Confluence / knowledge bases – connect workspaces and spaces
- Indexed web pages – add your public website or documentationIn these cases, you can start testing the AI agent the same day.

2. Standard Business Tools (A Few Hours to a Few Days)
For CRM, support, and project tools:
- HubSpot, Salesforce, Zendesk, Intercom, Jira, GitHub, Airtable, etc.
Most of the time is spent on:
+ Internal validation for access (security / IT)
+ Selecting which objects to sync (tickets, deals, contacts, projects…)
+ Defining access rules (which teams see what)
And in most cases, you can have a working MVP in 2–5 business days.

3. Custom Integrations (A Few Days to a Few Weeks)
For internal tools (homegrown apps, proprietary databases, custom APIs), we can build a custom connector:
- Connection via REST API, database, or middleware
- Definition of your business workflows (e.g. “if urgent ticket → create incident in Jira + notify on Slack”)
- Implementation of security policies and logging

Typically, a full project (from connection to first agents in production) is done in weeks, not months.

RAG stands for Retrieval-Augmented Generation. It’s an AI approach that allows the agent to use your up‑to‑date company data instead of relying only on what the model learned during pre‑training.

Why Is RAG Important for an AI Agent?
- Answers based on your real content
The agent uses your documents (Google Drive, SharePoint, Notion, Confluence…), tickets (Jira, Zendesk), emails (Outlook), indexed web pages, etc.
- Reduces hallucinations
Instead of “making up” answers, the agent grounds its responses in your actual data.
- Keeps answers always up to date
As soon as you update a process, an offer, or a policy, the agent’s next answer can reflect that change.
- Adapts answers to context
The agent can combine multiple sources (e.g. CRM + internal notes + product FAQ) to generate a contextual response.

How Does RAG Work in Practice?
1. A user asks a question in natural language
2. The agent retrieves the most relevant passages from your data (retrieval)
3. The language model generates an answer based on those retrieved passages (generation)
4. The agent can show the sources it used (documents, tickets, pages) for transparency

RAG is essential to build a reliable AI agent aligned with your data and business rules, instead of a generic chatbot.

Yes. We can add new connectors even if they don’t appear in the default list.

Different Ways to Connect Your Tools

1. Integrations via Model Context Protocol (MCP)
We can integrate new data sources via the Model Context Protocol (MCP).

This allows us to:
- Expose external systems and APIs as structured tools for the agent
- Securely retrieve and update data from your applications
- Quickly plug in internal or niche tools without building a full custom connector from scratch

If your system can be accessed through MCP, we can make its data available to the agent and use it in workflows, reasoning, and automation.

2. Classic data File Import
For simpler use cases, you can also provide:
- CSV / Excel files
- PDFs, Office documents, exported Google Docs/Sheets

These files can be indexed and used by the RAG system.

3. Custom API-Based Connector
If your tool has an API we can create a connector that allows the agent to:
- Read the necessary data (tickets, leads, projects, documents…)
- Create or update records according to your workflows
- Respect your permissions and security mode

Typical Process for a New Connector
- Quick analysis of your tool and its API / schema
- Definition of the use cases and data needed
- Implementation of the connector (read + write if needed)
- Configuration of permissions (which users/teams see what)
- Tests in a limited environment before broader rollout

Yes. Your AI agent can work even if your data is not perfectly “clean” or structured.
In fact, one of the main benefits of an AI agent is to make messy, scattered information usable.

What the Agent Can Handle Even with “Dirty” Data
- Unstructured documents: PDFs, slide decks, meeting notes, emails
- Multiple sources: Slack, Teams, Outlook, Notion, Google Drive, SharePoint, etc.
- Various formats: product sheets, internal procedures, support tickets, project docs, etc.
- Redundant or multiple versions: the agent can surface the most relevant and recent content.

How the Agent Helps Improve Data Quality Over Time
- Centralizes search across all your sources
- Surfaces the most relevant and up‑to‑date content
- Helps you detect inconsistencies (conflicting procedures, outdated docs…)
- Can be configured to ignore certain folders/spaces that are unreliable or obsolete

When Data Cleaning Becomes Useful
A full cleanup is not required, but it’s useful if:
- You have content that should never be used (retired offers, deprecated processes, expired legal docs)
- You want to enforce “golden” sources (official documentation, validated procedures)
- You operate under strict compliance rules (legal, healthcare, finance)

In those cases, we can help you define which sources the agent should prioritize or exclude, without requiring a complete data restructuring before you start.

Who can create and manage agents depends on how you structure roles and permissions in your organization.

Typical Roles Involved in Agent Creation
- Admins / IT Teams / Champions
+ Manage connectors and permissions (access to CRM, drives, internal tools, etc.)
+ Validate test and production environments

Operations / Business Owners / Knowledge Managers / Product Owners
- Define the use cases (customer support, internal support, onboarding, lead qualification, etc.)
- Configure business workflows (what the agent should do, when, and for whom)
Organize content sources (FAQ, knowledge bases, procedures)
Validate responses, examples, and tone of voice

Role-Based and Team-Based Control
You can define:
- Who can create or edit an agent
- Who can deploy an agent to production
- Who can use the agent in daily workflows
- Which agents are visible to which teams (Support, Sales, Customer Success, IT, HR…)

The goal is to let business teams design and own their AI agents, while IT and security keep control over permissions and connections to sensitive systems.

For data warehouses and databases, you usually don’t need a new specific connector.

We address these systems via our SQL query model, which allows the agent to:
- Run controlled, parameterized SQL queries on your warehouse
- Retrieve the data needed for analytics, reporting, or decision‑making
- Use query results inside reasoning chains and automated workflows

Because this is handled through our SQL layer, we don’t require a bespoke connector for each warehouse technology.

On top of automatic header and structure understanding of your documents, the platform can leverage a data catalog to enrich the context passed to the agent via RAG (Retrieval‑Augmented Generation).

Built‑In Data Visualization
On top of querying your warehouse, the agent can also provide data visualization:
- Generate charts and graphs (bar charts, line charts, pie charts, tables, etc.) from SQL query results
- Help you quickly explore trends, anomalies, and KPIs without leaving the interface
- Adapt visualizations to your question (“Plot this over time”, “Break this down by region”, “Show a comparison by owner”)

Users can search for information in natural language, get structured answers backed by SQL, and instantly see the data as visuals, making your database or data warehouse a true interactive analytics layer powered by AI.