Agents · create, run and govern

AI agents that work under clear rules.

Create, run and keep track of the company's agents in one place. Set each agent's owner, permissions and budget, see a record of everything it did and connect the agents to the same memory in the Brain.

Product screen, with sample data

Create

Start from the goal, not the settings.

Describe what the agent needs to do. The AI proposes the type, the instructions, the autonomy level and the routine, and a person reviews it all before the agent is created.

  • Create from a goal, with an editable proposal
  • Ready-made agent templates for a faster start
  • Export and import agents
  • A simulator to test before publishing
Product screen, with sample data

Run

Routines, channels and a view of the day.

Agents work at set times, when another system reports that something happened or when someone calls them. The Today tab shows what ran, what failed, how much it cost and what is waiting for approval.

  • Routines on a schedule or on an event
  • Channels: Slack, Microsoft Teams, Google Chat, Telegram and WhatsApp
  • Tools connected through MCP servers
  • A calendar and a board of routines and tasks
Product screen, with sample data

In processes

Agents that work inside the company's processes.

An agent can be a step in a process: it gets the task with the case data, returns the result and the process moves on to the rule or to the person who approves. The same rules, the same budget and the same record apply in there.

  • The agent gets the task with the case data and writes the result to a field
  • If the agent asks for help, the case waits for a person, moves on or closes
  • Every pass of the agent is on the case timeline
See Intelligent Automation
Product screen, with sample data

Govern

Owner, approvers and per-tool rules.

You set each agent's owner and who can approve its actions. Tools are the actions an agent carries out in other systems, such as looking up a record or sending a message. For each one you decide: allow, ask for approval or block.

  • Owner and approvers set per agent
  • Per-tool rules: allow, ask for approval or block
  • A notice to the approver on Slack, Google Chat, Telegram or WhatsApp
  • The option to mask personal data in logs and answers
Product screen, with sample data

Cost and audit

Every run leaves a record and has a spending cap.

The history keeps every run with its cost, duration and origin. The step-by-step view shows the question, each tool called and why the run stopped. The budget sets how much the agent can spend.

  • A monthly budget with a warning and an automatic stop
  • History with the cost, duration and origin of every run
  • A step-by-step view of every run
  • An audit of configuration changes
Product screen, with sample data

Catalogue and versions

Every agent in the company, versioned and tested.

The catalog shows each agent with its type, status and month spend, including AI routines that run outside the platform. You can keep changes in a draft, test them in the simulator or in regression tests, and only then publish.

  • Catalog with each agent's status and spend
  • External agents: AI routines running elsewhere (Claude Code, n8n, scripts) with the cost and summary of every run
  • Team view, with delegation between agents
  • Draft kept apart from the published version, with regression tests
Product screen, with sample data

Agent teams

See who delegates to whom.

The map shows agents organised into teams. A coordinator receives the work and splits it among the specialists, and every agent keeps its own rules and budget.

Product screen, with sample data

With the rest of the platform

Agents with memory, and with limits.

The Brain gives agents what the company knows. Intelligent Automation puts the agents into the company's processes. LLM Management decides which models they use and who can talk to them.

Brain

The memory agents read, with an owner, verification and audience on every article.

See the Brain

Intelligent Automation

Processes with agent steps, rules and approvals by people, designed from the company's procedure.

See Intelligent Automation

LLM Management and Chat

LLM Management handles the model providers and cost, and the Chat is where the team talks to the agents.

See LLM Management

FAQ

About the Agents

What kinds of agent can I create?

Internal assistants with different roles (assistant, specialist, analyst, operator and coordinator) and customer service agents. All of them can use the Brain as their memory: you choose what each one consults.

Can an agent act without anyone seeing?

It depends on the rules you set. Every internal assistant has an autonomy level, and every tool has its own rule: allow, ask for approval or block. Every action goes into the history, approved or not.

Can agents take part in processes with steps and approvals?

Yes. In Intelligent Automation, a process brings together steps done by agents, rules that pick the path and approvals by people, with a deadline. Every case is recorded with its timeline, and the approver gets the decision in the Approvals inbox and as a task.

Can I track an AI routine that runs outside Kluco?

Yes. Register the routine as an external agent: it keeps running where it is (scheduled task, script, Claude Code, n8n) and reports every run, with cost, summary and errors. The catalog shows whether it is on time, late or failing, and the month budget can warn or pause the routine.

How do I control what agents cost?

You set a monthly budget for each agent. The agent warns you before it reaches the cap and stops automatically when the limit is hit. The history shows the cost of every run.

Does personal data show up in the logs?

You can switch on masking of email, CPF, CNPJ, phone and card numbers in logs and answers, agent by agent.

Do agents work with any model?

Agents use the providers set up in LLM Management. To act in other systems, an agent needs a model that can use tools. Today that covers the models from OpenAI, Anthropic, DeepSeek, Grok and OpenRouter.

Agents

See the agents in action.

A live demo, from creating an agent to the record of every action.