What is Syntrum

Syntrum is digital workforce software. You don't build automations in it — you hire AI employees, called agents, and put them to work inside the tools your company already uses. Each agent has an identity, a job, and the judgment to do that job without being told every step.

This page explains the idea behind the product, how it's different from the tools you may be comparing it to, and the loop that makes an agent an agent.

The thesis: hire, don't build#

Traditional automation asks you to become a builder. You map out every step, connect triggers to actions, and account for every branch — and the moment reality drifts from your flowchart, it breaks and you go back to editing.

Syntrum asks you to become a manager instead. You describe the role, the objective, and the responsibilities, and the agent figures out the how. When something is unclear or risky, it asks — like a good employee would — rather than failing silently or guessing.

That shift changes what your job looks like:

How Syntrum differs from RPA and chatbots#

It's easy to file Syntrum next to tools you already know. Here's why it isn't those things.

Not RPA / workflow automation#

Robotic Process Automation and workflow builders execute predefined scripts. They repeat the exact steps you recorded, in the exact order, and they can't handle anything you didn't anticipate. A Syntrum agent isn't following a script — it observes what's actually happening, decides what to do based on its role and memory, and adapts when circumstances change. There's no flowchart to maintain.

Not a chatbot#

A chatbot waits for you to type, answers, and forgets. It's reactive and stateless. A Syntrum agent is proactive and continuous: it watches your tools for work, acts on its own initiative, remembers what it learned, and reports back — even when you're not looking. You can chat with it (see Chatting with agents), but conversation is how you guide the worker, not the whole product.

The short version#

The loop at a glance#

Every agent runs the same five-phase cycle, continuously:

Observe → Understand → Decide → Act → Report

• Observe — watch connected tools for new events.
• Understand — interpret them using role, memory, and company context.
• Decide — choose an action, checked against your guardrails.
• Act — do it, or pause to ask when a rule requires approval.
• Report — record what happened, why, and the outcome.

This loop is the reason an agent can be trusted to work unsupervised within limits you set. The full walk-through, with a worked example and where you can step in, is in The agent loop.

Humans stay in control#

Autonomy without control isn't useful. Syntrum keeps you in the loop through three mechanisms that work together:

• Supervision modes — how independent each agent is by default. See Supervision & modes.
• Guardrails — the workspace-wide floor for what any agent may do, including which actions are always blocked or always gated. See Guardrails & sensitive actions.
• Approvals — the queue where an agent asks before doing something your rules gate. See Approvals.

These resolve through one cascade — Workspace → Agent → Tool — where a block at any level is a floor that lower levels can't loosen. That mental model is The three planes.

What's next#

• The agent loop — the five phases in depth, with an example.
• Agents, workspaces & integrations — the object model in plain language.
• Memory & learning — how agents get better over time.