Skip to content
AI agents · Gaspa Brussels

Custom AI agents. Useful actions. Explicit limits.

An AI agent can consult sources, use tools and carry out sequences of actions. Gaspa defines its role, permissions and approvals to integrate it into a process your teams can monitor and take over.

5 stars4.96/5 on Sortlist

First 30-minute conversation · Free · No commitment

Gaspa symbol in blue glass
250

completed projects

4.96/5

Sortlist · 18 reviews

Your starting point

An AI agent needs to know what it can do.

Understand the request

The agent gathers context and useful information before suggesting the next action.

Prepare the work

Search, qualification and drafting can follow one another within authorised tools.

Submit or execute

Selected actions are carried out or presented to a person according to their sensitivity.

Understand what was executed

Logs and statuses make the tools used, errors and decisions requiring team intervention visible.

THE PROCESS AT A GLANCE

  1. 01

    Understand

    The agent gathers context from authorised sources.

  2. 02

    Propose

    An action is prepared with the elements to check.

  3. 03

    Approve

    A person authorises the sensitive action or takes over.

An approach tailored to your context.

Prepare the work. Keep the decision.

Imagine an agent helping process an incoming enquiry. It searches for context, gathers missing information and suggests an action. The team approves external messages or sensitive changes before execution.

Authorised toolsPermissionsApprovalTraceability

Understand

The agent gathers context from authorised sources.

Propose

An action is prepared with the elements to check.

Approve

A person authorises the sensitive action or takes over.

Illustrative example · AI agents This scenario does not describe a client project.

Discuss my situation
How we support you

From the agent’s role to observable execution.

Each phase produces concrete deliverables. You see progress, understand the choices and approve the relevant points before moving forward.

Choose the right use

01

We review the work to improve, data, access and useful approvals.

✓A priority scenario and evaluation criteria.

A role and permissions

02

The objective, accessible tools, authorised data and prohibited actions are documented.

✓Explicitly authorised access and actions.

Controlled journeys

03

Handling approvals, errors, execution limits and manual takeovers.

✓Approvals and manual takeover planned.

Observable evaluation

04

Test scenarios and action logs to understand results and correct behaviour.

✓Tests and logs to understand the agent.

Review with you

05

Key scenarios, errors and exceptions are examined before the next phase.

✓Visible approvals and discrepancies addressed.

Run the pilot and adjust

06

Quality, errors, costs and feedback are monitored to decide the next steps.

✓An observable use and documented limitations.

Making the right choices

An autonomous agent is not always necessary.

The choice depends on your users’ needs, the existing system and the cost of future development. We make the trade-offs clear before development begins.

A deterministic workflow

01

For a known sequence of steps, explicit conditions and rules make control easier.

✓Prefer simplicity when it meets the need.

An agent for variable contexts

02

The agent chooses some steps according to the request and available tools, within defined limits.

✓Evaluate decisions, not just answers.

Collaboration

From the first trial to supervised use.

A shared framework, demonstrations and explained decisions. Your team knows the next steps, expected deliverables and points to approve.

An approachable team

You know who manages, designs and develops your product. The right people join the conversation at the right time.

Clear checkpoints

You see progress regularly and approve screens, priorities and releases before the next step.

Decisions explained

Before taking action, we explain the impact on your budget, timeline and the product’s ability to evolve.

Step(01)
Discovery workshop

Choose

The use case, permitted actions and how to measure the pilot’s quality.

WHAT WE DO

  • Observe current work
  • Choose a specific use case
  • Review authorised sources
  • Define permitted actions
  • List errors to test
  • A role and a scope
  • Sources and access
  • Evaluation cases
  • Human approvals
  • Criteria for continuing

People involved

Project leadershipProject managerProduct DesignerYou and your stakeholders

What you approve

The use case, permitted actions and how to measure the pilot’s quality.

Key decisions

Supervision is part of the agent.

These decisions are reviewed with your team. They determine the possible uses, responsibilities and conditions for the next step.

(01)

Minimum permissions

Authorised tools, data and actions are defined separately. The agent does not receive all of a user’s access rights.

(02)

Approval and human takeover

Sensitive operations require approval. A person can interrupt the journey or continue it manually.

(03)

Logs and execution limits

Actions, errors, duration and resource consumption are observable. Limits prevent endless sequences and bound retry attempts.

Client reviews

Collaboration described by the people who experience it.

Our clients share what matters during a project: a team that is available, responsive and committed.

Read client reviews on Sortlist

Review 1 of 3

Alba
We greatly appreciated the team’s availability and responsiveness, as well as their sincere commitment to understanding the specific needs of golf and meeting our requirements as closely as possible.
Alba Golf
Alba Golf

Valentin Outters

Alba Golf

ProtectU
An excellent experience with Gaspa Agency! A big thank you to the team for their professionalism and efficiency. The application was delivered on time, with a real ability to adapt and, above all, constructive suggestions.
ProtectU
Reda Belyazid

Reda Belyazid

ProtectU

Cupigo
We appreciated Guillaume’s availability and the way his colleagues listened to us throughout our project.
Cupigo
Cupigo

Jules Gerard

Cupigo

Start a project

Let’s start with a precise, measurable role.

No specification needed. Tell us about your idea, your current tool or the problem you want to solve. We will help you work out the next steps.

Free first conversationNo commitmentConfidential enquiry

4.96/5★★★★★

Sortlist

250

completed projects

Guillaume Gasparini

Shall we discuss your project?

Guillaume Gasparini

Founder of Gaspa

Before suggesting a solution or a team, we take the time to understand your needs, constraints and project goals.

Schedule a meeting
“The team’s availability and responsiveness.”
Valentin OuttersHead of Product — Alba Golf AI

A few details to help us respond.

Fields marked with an asterisk are required. A few lines are enough to get started.

Our team replies to you personally.

Your information stays confidential and is never sold.

Frequently asked questions

AI agents: answers to help you move forward.

Yes, if those actions are within scope and access allows them. Confirmations, authorised recipients and logs are defined according to each action’s possible consequences.

Products and brands designed to evolve.

Axiom, Cupigo and Al Por Tres illustrate our work on product journeys, interfaces and identity. Expertise brought together to shape your project.

Axiom
01Gaspa project showcase
Axiom
SaaS platformVisual identity

Axiom

A digital ecosystem bringing together learning, content, live sessions and community.

Cupigo
02Gaspa project showcase

Cupigo

Mobile applicationProduct Design

Cupigo

A dating experience designed around real, time-limited interactions.

Al Por Tres
03Gaspa project showcase

AL POR TRES

Visual identity

Al Por Tres

A premium, welcoming identity for a brasserie within a padel club.