Custom AI
We design, build and integrate custom AI for companies: agents, process automation, software, data analysis and integrations with your systems. AI does the repetitive work, so your team's time goes to customers and the business. As a technology partner, we build on the HUB, Xarxalia's own platform, with roles, permissions, a log of every action and human oversight.
When it makes sense
If you need to cover a specific function, such as handling enquiries, sales, bookings or billing, MIA's digital employees already do it and your team can look after more customers. If the problem spans an entire process, several systems, large volumes of information or a tool that doesn't exist yet, that calls for a custom project.
| Criterion | MIA | Custom AI |
|---|---|---|
| What it is | Our own product, with digital employees already built | A solution designed and built for your case |
| When it fits | Specific functions: customer service, sales, back office, bookings, billing, marketing, voice | Entire processes, several systems, large volumes of information or a new tool |
| Roll-out | Target: five working days | In phases; scope and duration are set in Discovery |
| Integration | Everyday tools: Google Workspace, Microsoft 365, WhatsApp Business, CRM, billing | Corporate systems: ERP, CRM, databases and APIs, with SSO, MCP and A2A |
| Built on | The HUB | The HUB, plus whatever needs building on top of it |
What we build
Agents that read incoming information, decide the next step and act inside your systems: they open a ticket, update a record or draft a reply. Each agent has per-action permissions and an autonomy level, and escalates to a person any decision that isn't its to make.
Workflows connect teams, systems and data, with conditions, wait steps and approvals, and leave an audit trail of every step. What should be predictable is handled with rules; AI comes in where a document, an email or a request needs interpreting.
We build bespoke software with artificial intelligence at its centre: internal tools, platforms and digital products. If the tool already exists, we embed the models in it instead of replacing it.
Working from the organisation's data, models analyse, classify, predict and recommend. Each result feeds into a specific decision, such as which order to prioritise or which customer to call first.
We connect AI to the organisation's ERP, CRM, Microsoft 365, Teams, Google Workspace and databases, through APIs or via MCP and A2A. Staff can sign in with corporate SSO, and their existing roles and permissions still apply.
Describe the problem and we'll tell you whether a custom project makes sense, whether MIA is enough, or whether it isn't worth it yet.
How we work
In Discovery, once we understand the problem, we set the scope and duration of the custom AI project. The solution doesn't go into production until it has been tested in a pilot.
Discovery
We analyse the process, the systems involved and the data available, and set the scope, the expected outcome and the duration.
Architecture
We design how the models, agents, integrations and permissions fit together, which part the HUB covers and what needs to be built.
Pilot
We put the solution to work in a limited scope, with real cases and supervision, to check that it solves the problem.
Integration
We connect the solution to the ERP, the CRM and the other systems involved, using the organisation's SSO and roles.
Production
The solution goes into production across the agreed scope, at the autonomy level the organisation chooses. Every action is logged.
Evolution
We review the results and adjust the solution. When you decide to, we extend it to other processes.
Integrations and protocols
Staff keep working with their usual tools. The AI connects to them through APIs or open protocols and uses the organisation's own access rights and permissions.
Systems
Protocols
Identity and control
MCP (Model Context Protocol) is an open standard for connecting AI models and agents to tools and data sources through a common interface, without writing a separate integration for each combination.
A2A (Agent2Agent) is an open protocol that lets AI agents from different systems and vendors communicate, share out tasks and coordinate their work.
Architecture examples
These are conceptual examples of how the pieces fit together; they don't describe any client.
A company receives thousands of documents, enquiries and data records every month by email and web forms. The AI reads them, extracts what matters and cross-checks it against the ERP and the CRM. Within the permissions set, it proposes or carries out the action and reports the result; anything that isn't its call goes to a person.
Inputs
AI
Systems
Output
Staff ask their questions in Teams, by email or on the intranet, and the assistant looks for the answer in the document management system, the intranet, the CRM or the ERP, always within the permissions of the person asking, which come through corporate SSO. It answers citing the source, prepares a draft when needed and passes whatever it can't resolve to the person in charge.
Inputs
AI
Systems
Output
Governance
The data the HUB processes is hosted on servers that Xarxalia administers in Barcelona, inside the European Economic Area, and each organisation works in a separate space. If the project uses other providers, such as a third-party AI model or the organisation's own cloud, they appear in the sub-processor register. You can request it before signing, along with the data processing agreement (DPA), the security policy and the service level agreement. Every action is logged with a date and an owner, and the organisation decides how much autonomy the AI has, always with human oversight.
Technology, security and governance →What problem should artificial intelligence solve in your company?
Before we talk about technology, we want to understand the problem. Tell us which process you want to transform, which systems are involved and what result you need.
A custom AI project is an artificial intelligence system that Xarxalia designs, builds and integrates to solve a specific problem for an organisation: an entire process, several connected systems, large volumes of information or a tool that doesn't exist yet. It can include agents, process automation, software, data analysis and integrations with the ERP, the CRM or Microsoft 365. It is built on the HUB, Xarxalia's own platform, with whatever the case needs developed on top.
The duration of a custom AI project depends on its scope, and Xarxalia sets it in Discovery, the first phase, once it is clear which process needs solving, which systems are involved and what data is available. That's why we don't give a figure until we understand the problem. Before reaching production, the solution goes through a pilot that checks it meets its goal.
MIA is the right fit when the problem is a specific function already covered by one of its seven digital employees, such as reception, sales, bookings or billing. The target is to have MIA live in five working days once Xarxalia has access to the relevant systems and the information it needs about the business. A custom AI project is the right fit when the problem spans a whole process, several corporate systems, large volumes of information or a tool that doesn't exist yet, and Xarxalia builds it in phases, with a pilot before production. Both routes share the HUB, Xarxalia's own platform, and the same rules on data and human oversight.
MCP (Model Context Protocol) is an open standard for connecting AI models and agents to tools and data sources, and A2A (Agent2Agent) is an open protocol that lets agents from different systems communicate and coordinate tasks. They matter because they are open standards: they reduce dependence on a single vendor and make it possible to add new tools or agents without coding each connection from scratch. Xarxalia works with MCP and A2A in its custom AI projects, alongside APIs and corporate SSO.
Yes: Xarxalia integrates custom AI projects with the organisation's ERP, through connectors or APIs, and with Microsoft 365 and Teams, so staff work with the AI from the tools they already use. In Discovery, the first phase, we review what each system allows and decide how to connect it. Users can sign in with corporate SSO, and access respects the roles and permissions the organisation defines. We also work with CRMs, Google Workspace and databases, and with the open protocols MCP and A2A.
In a custom AI project, the data the HUB processes stays on servers that Xarxalia administers in Barcelona, inside the European Economic Area. If the project uses other providers, such as a third-party AI model or the organisation's own cloud, they are listed in the sub-processor register. Each organisation works in a separate space, users can sign in through corporate SSO under the roles and permissions the organisation defines, and every action is logged with a date and an owner. Before signing you can ask Xarxalia for the data processing agreement (DPA), the security policy (encryption, access control and two-factor authentication) and the sub-processor register.
Before asking Xarxalia for a custom AI project, be clear about which process or challenge you want to solve and what result you need to achieve. If you already know which systems are involved, such as the ERP, the CRM or Microsoft 365, say so. On the contact form you also choose the scope, from a single process to the whole organisation, and can optionally add an indicative budget and the documents you'd like to receive. The next step is Discovery, the first phase, where we analyse the problem, the systems and the data, and set the scope and duration. If what you need is a specific function that MIA already covers, we'll tell you.