Layer 04 · AI integrated with your systems
Your ERP, CRM, document management, core systems and BI already run the business. We do not replace them. We add layers so your team can ask questions, cross-reference data across systems and act with AI, with the same permissions and the same control you have today.
Example animation with fictitious data: records from your ERP, CRM and document system are connected without migrating them, the query goes through identity, permissions and logging, and an agent answers by combining the ERP ship date, the CRM customer and the contract.
How it connects
Your systems stay at the base, as they are. On top goes an integration layer, which defines how AI reaches each system. Next is a governance layer, which decides what AI can see and do. At the top are the interfaces where your team talks to AI and the agents that do the work.
Each layer can grow without touching the others: add a system, switch assistants or open a new action.
Where your people ask and decide: your intranet, Teams or WhatsApp, your company AI assistant, or agents that work on their own.
Who can see what, which actions are allowed, what needs approval and what gets logged.
The paths AI uses to read from and act on each system. You pick one per system, based on what that system already offers.
Nothing is replaced or migrated. They remain the source of truth, with their own data and permissions.
Integration layer
No single path fits everything. The choice depends on what each system already offers and what you want AI to do in it. Most companies end up with a combination.
An open protocol that lets any compatible AI assistant discover and use a system's functions, with each person's own access.
BEST WHENSeveral assistants or agents will use the same system, and you want to control access from one place.
Packaged instructions and commands the agent runs step by step. They are versioned and reviewed like any other procedure.
BEST WHENInternal, repeatable, high-volume tasks where speed and cost per operation matter.
The interfaces your system already exposes. They are often the foundation the other paths are built on.
BEST WHENThe system has a stable, documented API. It is the most direct path.
The system signals when something happens ("an invoice arrived," "a status changed") and the AI reacts, with no constant polling.
BEST WHENThe work starts with a business event. This is the foundation for agents that work on their own.
Off-the-shelf connectors for commercial systems. They save time, but you need to review the permissions they come with.
BEST WHENYour system is a commercial product with mature connectors.
The agent uses the system's screens the way a person would: it reads, searches and fills in forms.
BEST WHENLegacy systems or external portals with no API. It is a bridge: slower, and always supervised.
The conversation lives inside the system your team already uses and respects its permissions.
BEST WHENMost of the work happens in a single system and there is no need to cross-reference others.
Read-only access to the database or data warehouse, ideally on metrics your team has already defined.
BEST WHENQuestions about metrics and analysis. Nothing is changed, and the numbers come from agreed definitions.
What it looks like
Pick a role and ask the question. You will see which system the AI queries, through which path and with what result. Before it does anything, it asks for your approval.
The systems, paths and data in this example are illustrative. Yours depend on the assessment.
EXAMPLE · FICTITIOUS DATA
Which key accounts have overdue invoices and contracts up for renewal this quarter?
Governance layer
The governance layer sits between AI and your systems. It stays the same no matter which integration path or assistant your team uses.
AI acts with each person's identity and sees only what that person can already see in each system. No parallel access is created.
Every query goes through one central point (an AI gateway) that enforces the rules, instead of loose credentials for each agent.
Who asked what, which system was queried, what was answered and what was done. All of it is available for audit.
Looking things up can be immediate. Creating, approving, sending or paying requires explicit confirmation from a person.
Each integration exposes only what is needed. You set caps, schedules and data that never leaves the system.
Tested before going to production and reviewed in use: answer quality, errors and unexpected queries.
How we start
We review which systems you have, what each one offers (API, events, connectors or screens only) and which questions are hard for your team to answer today.
We connect one or two systems for lookups, without changing anything. Your team tests with real questions and their own permissions.
We enable the first actions (create a task, prepare a draft), always with a person's confirmation and a log.
We add systems, departments and, where it makes sense, agents that work on their own. Governance rules are not rebuilt: they are extended.
With the rest of Appropia
If you use our document repository, LuxFlow or Cortex, AI queries them through the same paths and under the same governance as your ERP or CRM. No special route, no separate governance.
AUTONOMOUS AGENTS
Here, a person asks and the AI answers. When a process needs an agent working continuously, AgenticOS uses these same integrations and the same governance: permissions, limits, approval and a log of every action.
Explore AgenticOSFrequently asked questions
No. Your ERP, CRM, document management, core systems and BI stay as they are and remain the source of truth. What we add are layers: an integration layer, a governance layer and, on top, the conversational interfaces and the agents.
It depends on what each system offers. If it has a stable API, we use that API. If several assistants will use it, the open MCP standard is a good fit. Commands (CLI) work for internal, repeatable tasks. If the work starts with a business event, we use events. If there is no API, the agent works through the user interface. It is almost always a combination, defined during the assessment.
Model Context Protocol is an open standard for connecting AI assistants to systems and data. Its advantage: you build the connection once and any compatible assistant can use it. It is one integration path, not the only one.
There are alternatives. The agent can operate the system's screens the way a person would, or read the data in read-only mode. It is a useful bridge for legacy systems, but it is slower and more fragile, so it is supervised and replaced with a more stable path when one exists.
Only the actions that are explicitly enabled, and only after a person approves them. We recommend starting read-only and opening actions one at a time.
AI acts with the identity of the person asking and inherits that person's permissions in each system. If someone has no access to a contract or a customer, the AI has no access to it either when answering that person. Every query is logged.
We design the integration and governance layers so they do not depend on a single model. Switching models or adding another assistant does not mean rebuilding the connections to your systems.
Here, a person asks and the AI queries your systems to answer or propose an action. In AgenticOS, agents work autonomously and continuously. Both use the same integrations and the same governance.