It does
- 01
Read your real data, with the rights it was given.
- 02
Propose one precise action, in plain words, before acting.
- 03
Execute what you approved inside your tools, and trace it.
AI agents
Agents that act,not agents that chat.
A Digix agent reads your data, proposes, waits for your approval when it matters, then executes inside your tools. Everything is logged.
01 — Console
Pick an agent. Watch it work.
Project-manager agentWhatsApp · ERP
- Client
Can you move the delivery to Thursday?
- Tool
schedule.read(project=“Atlas site”)
- Tool
proposal → project manager: “Thursday 2 pm?” · approved
- Tool
task.move(Thu 14:00) · team.notify()
- Agent
Done. Delivery is Thursday at 2 pm and the team has been told.
02 — Guardrails
How an agent is built.
Five steps, always in this order. It is what makes an agent usable in a real business.
- Request
- Reading
- Proposal
- Approval
- Log
- 01 — Request
- A message, an email, an event in your tools.
- 02 — Reading
- The agent reads your real data, with the rights it was given.
- 03 — Proposal
- It proposes one precise action, in plain words.
- 04 — Approval
- A human approves what matters. The rest goes through on its own, by your rules.
- 05 — Log
- Every action is traced: who, what, when, on what basis.
Log · one action, from message to trace
- 10:41:02Message received · WhatsApp · client “Atlas”
- 10:41:03Read schedule.project(“Atlas site”) · rights: read
- 10:41:05Proposal: move the delivery to Thu 14:00
- 10:52:17Approved by Yassine · from the ERP
- 10:52:18task.move(Thu 14:00) · team.notify() · ok
- 10:52:18Reply sent to the client · log written
03 — Limits
What it does. What it never does.
Never
- 01
Act on a payment or a contract without human approval.
- 02
Read data the person asking has no access to.
- 03
Make up an answer when it has no source.
07 — Contact
Let’s talk about your project.
Leave your email and we come back with a plan, not a brochure.
or call us