Created in September 2025 within the Accelis Group, GAIA Labs designs “AI colleagues” that IT service providers deploy for their clients. The lab makes control its main selling point: each agent works within a framework that the provider or the IT department (DSI) sets client by client. This content is sponsored by GAIA Labs.
Based at the group’s technology campus in Seine-et-Marne, the lab targets MSPs (managed service providers), ESNs (digital service companies), and integrators. Its approach stems from an observation: in every company, a portion of time is spent on repetitive tasks, such as following up, verifying, copying, or sorting. GAIA Labs wants to entrust this repetition to an artificial intelligence agent, leaving tasks requiring judgment to people. The lab believes that trust precedes delegation, and therefore primarily targets IT departments, which want to know at all times what an agent can read, what it can do, and how much it costs.
A colleague assigned to a manager
An AI colleague integrates into the team like a new hire. It is assigned to a manager and has its own email address, access to instant messaging or Microsoft Teams, a calendar, and a private workspace where it produces dashboards, PDF summaries, and case tracking. Nothing is installed on workstations: the agent joins the tools already in place.
Its onboarding follows three steps. A 30-minute scoping session with the manager and the IT department defines its missions, access rights, and spending limit. On the first day, the agent is in read-only mode: it prepares its first drafts within the hour, and nothing is sent without validation. This is followed by supervised autonomy governed by the rules set initially. The catalog includes ten operational business programs, including management controller, SOC analyst, level 1 IT support, and business developer, as well as three programs in pilot phase.
A console to configure each agent
The main difference lies in the GAIA console, where the provider or IT department sets the framework for each agent, for each client, using five levers. Permissions are granted application by application, with five levels: read, create, modify, delete, and action. The connectors to email, files, ERP, or ticketing tools are displayed with their expiration date. The pace is controlled by call and item read limits, and the budget is set in euros per agent per client. Finally, the intelligence level is chosen from four packs, Essential, Balanced, Executive, and Apex, which determine both the agent’s capabilities and its cost.
Regarding costs, the rule is simple: at 80% of the monthly limit, the manager and the MSP are notified, and at 100%, the agent stops and remains paused until the next month. The lab specifies that there is no billable overrun.
No credentials in the hands of the agent
The security principle is established from the start: the agent holds no credentials. Access to applications is entered by the MSP in the GAIA data gateway, which itself opens the sessions. The agent formulates a request, the gateway checks its permissions, performs the operation, and returns only the result. Every request is logged, including when it is denied. For agents in service within the Accelis Group, the gateway refuses about one in twelve requests.
Data is hosted on servers in France, and business data remains in the client’s applications. Under GDPR, the client company remains the data controller, and the group acts as a processor. The lab advises starting with read-only access and recommends against granting delete permissions, a precaution it says aligns with ANSSI recommendations for generative AI systems.
The GAIA Labs partner program operates with no term commitment, no minimum volume, and no setup fees, and it is possible to start with just one agent. A 30-minute demonstration can be requested, with a response promised within 24 business hours.
