Custom agents

Your process does not fit a ready-made agent? UPX builds and manages one for you.

We design, integrate and operate tailor-made AI Agents for processes that involve proprietary systems, specific rules and sensitive data, with the governance of a company that has been doing cybersecurity for more than 20 years.

Companies that trust UPX

  • Bradesco
  • Nubank
  • BTG Pactual
  • Totvs
  • Ascenty
  • Live!
  • G4 Educação
  • EVEO

When custom is the right callMost operations are served by an agent from the catalog. Custom exists for the cases where the process belongs to your company, not to your department.

Proprietary or legacy systems

Your ERP, database or internal tool has no ready-made integration.

Processes that cross departments

The flow runs through finance, operations and support at the same time.

Specific rules and compliance

Approval limits, sign-offs and restrictions a generic agent does not know about.

No in-house AI team

You want the outcome, not to build and maintain the technology.

How it works

Implementation starts by understanding the work that needs to be done, not by choosing technology. At every stage it is clear what belongs to your team and what belongs to UPX.

  1. 01

    We understand the operation

    We map the process, the rules, the systems involved and the expected outcome.

    Your team
    Shows the process and gives access to who knows it
    UPX
    Runs the assessment and documents the scope
  2. 02

    We configure the agent

    Skills, knowledge, integrations, channels, permissions and autonomy levels are defined for that operation.

    Your team
    Authorizes access and validates the rules
    UPX
    Configures the agent and builds the integrations
  3. 03

    We validate with your team

    The agent is tested within the real context before taking on the flow.

    Your team
    Tests with real cases and flags adjustments
    UPX
    Fixes, calibrates and documents the behavior
  4. 04

    We manage the operation

    UPX monitors and adjusts the configuration as the process evolves.

    Your team
    Tells us when the process changes
    UPX
    Reviews rules, integrations and permissions

The schedule depends on the complexity of the process, the integrations and the rules involved. Scope and timeline are defined after the assessment.

An agent with access to your systems is a privileged user

Controlling that kind of access — least privilege, human approval and a trail of every action — is what UPX has been doing for over two decades. The agent inherits that discipline.

Least privilege
The agent accesses only the systems and data released for that process. The scope is set in the configuration, not by the agent.
Four autonomy levels
Query, recommend, execute with approval, or execute autonomously. The level is set per action, so routine tasks can run on their own while sensitive operations still require human approval.
Traceability
Every action is logged: what was queried, what was proposed, when, in which system and under which permission. The history is available for review and audit.
Data protection
Operations follow Brazil's Law 13.709/2018 (LGPD). In AI Agent contracts, UPX acts as data processor; the legal basis remains with your company.
Your data does not train models
Content processed by Secure AI Agents is not used to train UPX models or third-party models.
In-house operation
The UPX SOC is internal and runs 24/7/365, with no outsourcing. Whoever operates is accountable.

Certifications and attestations

  • SOC 2 Type II
  • ISO 27001

UPX maintains SOC 2 Type II and ISO 27001, with independent audit over its information security controls.

We do not hand over the agent and disappear

The configuration stays under UPX management. As the process evolves, every piece can be reviewed and adjusted — without requiring an AI team on your side.

What stays under UPX management

  • Skills

    The capabilities the agent uses to carry out the tasks in the process.

  • Knowledge
  • Integrations
  • Channels
  • Permissions
  • Autonomy

Ready-made or custom?

The catalog covers the common processes of each department and industry, activated self-service. If your process is in there, there is no reason to pay for a project.

Comparison between the ready-made catalog agent and the custom agent, across five aspects.
Compared aspectReady-made agentCustom agent
Who it is forCommon processes of a department or industryProcesses specific to your company
IntegrationsReady-made connectors from the catalogIncludes proprietary and legacy systems
ConfigurationYou activate and adjustUPX designs and configures
ManagementSelf-serviceUPX manages continuously
ContractingMonthly credit tierPackage of hours

How contracting works

The custom agent is contracted by package of hours, sized according to the complexity of the project. There are two fronts: implementation and ongoing management.

Implementation

  • Process assessment, with scope and timeline defined at the end.
  • Configuration of skills, knowledge, integrations and channels.
  • Definition of permissions and autonomy levels per action.
  • Validation with your team before the agent takes on the flow.
  • Dedicated onboarding and implementation follow-up.

Ongoing management

  • Review of the configuration as the process evolves.
  • Adjustment of rules, integrations and permissions.
  • Monitoring of the operation by the UPX team.
  • Custom invoicing and billing terms.

The size of the package, what consumes hours and the renewal terms are defined in the proposal, after the assessment. Talk to a specialist to get the scope and price for your case.

FAQ

Frequently asked questions

What usually comes up before opening a custom agent project.

  • How long does it take for the agent to go live?

    It depends on the complexity of the process, the integrations and the rules involved. The assessment is what defines scope and timeline — and it happens before any commitment on dates.
  • Can I start with a ready-made agent and customize later?

    Yes, and in many cases it is the faster path. A catalog agent goes live self-service and shows where the process actually needs its own configuration. The custom assessment starts from a much better place when there is already real usage.
  • Where is my data? Does UPX use my data to train models?

    No. Content processed by Secure AI Agents is not used to train UPX models or third-party models. Operations follow Brazil's LGPD: in AI Agent contracts UPX acts as data processor, and the legal basis remains with your company. The retention policy is defined by contract and, once the contract ends, data is deleted within the agreed period.
  • What if my process changes?

    That is what ongoing management covers. The agent's configuration — skills, integrations, permissions, rules and autonomy levels — stays under UPX management and is reviewed as the operation evolves. You tell us the process changed; the adjustment is ours.
  • Do I need someone technical on my team?

    Not to build or maintain the agent. You need someone who knows the process, for the assessment and the validation, and someone who authorizes access to the systems. Configuration, integrations and maintenance stay with UPX.
  • Can the agent take actions on my systems on its own?

    Only at the autonomy level you define, and per action. There are four levels: query, recommend, execute with human approval, and execute autonomously within previously authorized rules. Routine tasks can run on their own while sensitive operations still require approval, within the same agent. Every action executed is logged for review and audit.
  • How much does a custom agent cost?

    Contracting is by package of hours, sized according to the complexity of the process, the integrations and the rules involved. Price and scope are presented in the proposal, after the assessment.
Start with the assessment

Tell us which process you want to hand to an agent.

Leave your contact details. A UPX specialist will reach out to understand the process and the systems involved.

Schedule an assessment

By submitting, you agree to our Privacy Policy.