Skip to content

Platform

Automation and AI agents (Roadmap) (Roadmap)

Workflow automation and AI agents are on the SyncedOffice roadmap. This page explains how they are designed to work inside each organization's workspace. (Roadmap) (Roadmap)

How are automation and AI agents designed to work? (Roadmap) (Roadmap)

The design was accepted in September 2026. Nothing described here runs yet. (Roadmap) The first slices each have an exit test they must pass before they count as built. (Roadmap)
  • Events from each workspace

    Changes in an organization's workspace become events on that workspace's own channel. A workflow in one organization's workspace cannot subscribe to another's.

  • Actions only through tools

    An agent acts only through named tools with permissions, never by reaching into systems directly. Each tool call is recorded, and a change waits for a person to approve it unless a rule allows it.

  • Knowledge stays in the organization's database

    Search over an organization's documents runs inside that organization's own PostgreSQL database, not in a shared index.

  • Tested before it ships

    Each agent ships only with scenario tests that fail when its behaviour breaks, and defences against instructions hidden in the content it reads.

Why build it inside each workspace?

Organizations should not share an agent's memory, its tools or its mistakes. Running each organization's workflows inside its own workspace keeps the same boundary the back office already has.

It also means an agent's record of what it did sits with the organization's other records, where the organization can review it. (Roadmap)

Which workflow would you hand over first? Tell us.

Join the pilot