Agents, workspaces & integrations

Syntrum has a small object model, and once it clicks the whole product makes sense. Three things do most of the work: workspaces, agents, and integrations. This page defines each in plain language and shows how they fit together.

The one-diagram version#

Everything lives inside a workspace. Agents belong to the workspace. Integrations are connected at the workspace and then handed to individual agents.

Workspace = one company#

A workspace is the container for one company (or one team, or one project). It holds its own agents, its own connected tools, its own knowledge, its own people, and its own billing. Nothing is shared across workspaces — switching workspaces is like walking into a different company's office.

• You can belong to multiple workspaces and switch between them with the workspace switcher in the sidebar. Syntrum reloads so every view reflects the workspace you're in.
• Creating a workspace starts a clean slate — empty of agents and data — and you're switched into it immediately.
• Your role is per workspace: you might own one and be a viewer in another. See Team & roles.

Agents = your AI employees#

An agent is a live AI worker that belongs to a workspace. Each agent has:

• an identity — a name and role, visible everywhere it appears;
• an objective and responsibilities — what it's for and what it owns;
• a communication style and heartbeat — how it reports and how often it checks in;
• its own memory — what it has learned and been taught;
• a supervision mode — how much it may do on its own.

Agents are created from templates — reusable blueprints. Design a template once (its role, objective, responsibilities, and behavior) and hire as many agents from it as you like. See Hire your first agent and Templates.

Every agent runs the agent loop continuously, and you manage them from the Agents list — starting, stopping, restarting, and opening each one. See Managing agents.

Integrations = per-user OAuth connections#

An integration is a connection to one of your company's tools — Gmail, Slack, Salesforce, and hundreds more. Integrations are the agent's senses and hands: without them an agent can think but can't see or act.

Two things are worth understanding about how integrations work:

Connected at the workspace, attached to the agent#

Connecting a tool authorizes it for the workspace — it becomes a shared, ready-to-use login. But a connected tool does nothing until it's attached to a specific agent at a chosen access level (Observe, Assist, or Act). Connection makes the tool available; attachment puts it in a particular employee's hands. See Connect a tool.

Per-user OAuth#

Connections are authorized through each provider's own consent screen — per-user OAuth. You sign in on the provider's page, grant scopes, and Syntrum stores and refreshes the resulting tokens (encrypted). Syntrum never sees your password, and the exact scopes granted are visible on the connected tool's card.

How they relate#

Putting it together with a concrete example:

1. Your workspace is Acme.
2. You connect Gmail and Slack as integrations — Acme now has those logins available.
3. You hire a Sales Follow-Up agent from a template.
4. You attach Gmail to that agent at the Assist level, so it drafts replies for your approval.
5. The agent runs its loop, observing Gmail, and its work appears on Home and in Activity.

The same Gmail connection could be attached to a second agent — say a Support agent — with different access, entirely independently.

The supporting cast#

A few more objects round out the model:

What's next#

• Memory & learning — what an agent remembers and how it improves.
• The three planes — how workspace, agent, and tool controls combine.
• Connect a tool — put the connected-vs-attached distinction into practice.