Skip to main content
Bonx AI agents handle tasks that are specific to your processes. You describe a task in natural language, attach it to a process step, and Bonx runs it with the platform tools. You decide what the agent does, when it starts, and which actions it can complete without approval. You can adapt each agent to your rules, data, and routines.

​
Composing an agent

An agent is defined by three things:
  • instructions, written in natural language, describing the task to carry out and the rules to respect;
  • a trigger, the process step at which the task runs (so the same record can fire different agents depending on the step);
  • a validation mode: by default every proposed change waits for your approval; you can switch to automatic acceptance once a task is safe enough to run on its own.
A single agent can carry several tasks, triggered at different steps of the same process.

​
What an agent can do

Agents have the same tools as the assistant: searching and reading your records, matching an article or a lot, creating and updating records, reading an attachment. In practice they take on tasks such as preparing an order from an incoming document, matching lines against your master data, or proposing a replenishment.

​
Following up and correcting

Every run leaves a trail: the agent’s messages, the proposed changes, your decision (accepted, rejected, cancelled, auto-accepted) and the records created. You can relaunch a run, and leave feedback that helps improve the instructions.

​
From the API

You can trigger an agent on a given context and follow its progress, to fold these automations into your own flows.
Agents act within the limits of the rights they are given, and their actions remain traceable.