Agents that work with your company's data
Create an agent from a ready template, choose what it can read and put it to work: it responds with the source of the number, runs on its own at the scheduled time and only acts after someone authorizes.
The flow you build, piece by piece
From trigger to delivery: the agent reads the data you allowed, checks documents and other systems, asks for approval before acting and delivers the answer with its source. And Flows chain all of it with pipelines and notifications.
Agent tools
Flows sit underneath everything: they trigger the agent, run pipelines and send notifications, in sequence.
Click a piece to see what it does.
Trigger
What it does
It is what puts the agent to work: someone asks in the chat, the scheduled time arrives, an event happens (a load failed, a metric went off track) or a Flow calls the agent in the middle of a sequence.
Example
Every Monday at 8 a.m., the executive agent prepares the weekly summary without anyone opening the screen.
Agent
What it does
The agent combines the instructions you wrote (its role, tone, what it may and may not do) with the chosen AI model. The model runs with your company's AI key — OpenAI, Anthropic or Google — so the model cost stays in your provider account.
Example
A Sales Director that only reads sales and targets, answers plainly and always shows where the number came from.
Gold layer data
What it does
The agent queries only the curated Gold layer tables you allowed, read-only. It does not read the source system's database or the raw layer, and the permission of whoever asks applies to every query.
Example
“How much did we sell per channel last month?” — the answer comes with the query used and the table it came from.
Documents and memory
What it does
Policies, discount tables, targets and manuals come in as documents the agent searches and cites. Memory keeps what the company taught along the way, so it does not ask again.
Example
When calculating margin, the agent checks the current discount table and cites the document.
External tools (MCP)
What it does
Through the MCP standard, the agent uses tools from other systems — CRM, support, spreadsheets. Each tool is approved before it joins the agent, and what it can do is recorded.
Example
The support agent opens the customer's history in the ticketing system before answering.
Specialist agents
What it does
A coordinating agent can ask specialists (sales, finance, operations) and combine the answers. Permission is rechecked at every hop: the specialist only sees what the person who asked can see.
Example
The Executive Analyst asks the Sales Director for sales and the Pipelines Monitor for the load status, and delivers a single summary.
Human approval
What it does
When the agent wants to ACT — rerun a pipeline, send a report, notify someone — it proposes the action in an approval card. Execution only happens after a person approves, and the decision is recorded.
Example
The Pipelines Monitor proposes rerunning only the failed step; the owner approves with one click.
Delivery
What it does
The result arrives as an answer in the chat, with the steps and the source of each number; as a report; or by email and WhatsApp. Every run is kept in the agent's activity, with cost and duration.
Example
The weekly summary arrives in the board's email and the manager's WhatsApp, with the link to check the numbers.
Flows
What it does
In Flows you connect boxes that run in order, like automation editors: when something happens, run a pipeline, ask the agent for a summary, check a condition and send the notification. Each step shows what it received and what it delivered.
Example
When the ERP load finishes, the Flow refreshes the Gold layer, asks the sales agent for the day's analysis and sends it by WhatsApp.
Agents that work with your company's data
Start from a ready template
Commercial, finance, marketing, executive, pipelines monitor — or describe what you want and the agent is built for you to review.
Response with the source
The agent queries only the data you allowed and responds with the number, the explanation and where it came from.
Runs on its own
Schedule: the weekly summary every Monday at 8am, the load checks every night.
Asks for authorization before acting
Reprocess a pipeline, send a report by e-mail or WhatsApp: the agent proposes, a person approves.
Company documents and memory
Discount table, targets, policies: the agent consults and cites. And remembers what the company taught.
Agent team and external tools
An agent can consult experts and use tools from other systems (MCP), each approved beforehand.
Visual automation that chains agents, pipelines and notifications
Build the sequence by connecting boxes, without coding: what triggers it, what runs, what the agent analyzes and who gets the notification.
Starts on its own
On a schedule or on an event: a load finished, another failed, a metric went off track.
Runs pipelines
One box refreshes the Gold layer before the agent looks at the numbers.
Calls agents
Ask the right agent for the analysis, the summary or the comparison, in the middle of the sequence.
Picks the path
Conditions split the path: notify only if the target was missed, ask for approval only if action is needed.
Notifies who needs it
Email, WhatsApp or report — and every run is recorded, step by step.