How it works
Rete.si runs one loop. Your agents do their work, the good parts become playbooks, a person approves them, and the playbooks that help spread to the rest of the team.
Demo video coming soon
Until then, the six steps below show the whole loop.
The playbook loop
-
Step 1: Agents work
Your agents handle real tasks: tickets, chats and requests, using the tools and playbooks you gave them.
You choose each agent's role, model, tools and playbooks.
-
Step 2: Private data is removed
Before any AI model reads the work to remember or learn from it, emails, phone numbers, card numbers, IBANs, addresses, keys and any terms you list are replaced with placeholders like [EMAIL_1].
You add your own terms to mask, per workspace or per agent.
-
Step 3: A playbook is proposed
After a finished task, the agent looks back at what worked and proposes a short playbook other agents could follow.
Team playbooks are off until you turn them on, and each agent can be opted out.
-
Step 4: Safety check
Every proposal is scanned for prompt injection, attempts to read credentials, remote code and other risky instructions, and scrubbed again. Approving stays locked until the check passes.
You see the result and the reasons before you decide.
-
Step 5: A manager approves
A person reads the playbook and approves it or rejects it with a reason. Nothing is used by any agent before that.
Every decision is written to the audit log.
-
Step 6: It spreads, results are tracked
An approved playbook goes to the agent that proposed it and, if you choose, to every agent on its team. Rete.si records how often it is used, when agents flag it as unhelpful, and how tasks went with it and without it. Playbooks that stop helping are retired.
You decide what spreads and what is retired.
How you know a playbook helps
-
Used
How many times agents loaded the playbook while they worked.
-
Flagged
How many times an agent marked it as unhelpful for the task in front of it.
-
With and without
Finished tasks and time to finish, for tasks that used the playbook compared with the same agents' tasks that did not. Shown only once there are at least 5 tasks on each side.
These numbers come only from your own workspace. When a playbook stops helping, a manager retires it with a reason; it stays in the history but no agent uses it again.