SyntrumDocs
Open app

Hire your first agent

Hiring an agent is like onboarding a new teammate: you give it an identity, tell it what it's responsible for, decide how it should communicate, and set how often it checks in. Syntrum offers two ways in — a quick Hire New Agent flow that starts from a template, and a guided builder that designs a template from a few questions. This page covers both, plus what happens the moment you deploy.

Templates vs. agents. A template is a reusable blueprint — an agent's role, objective, responsibilities, and behavior. An agent is a live worker created from a template. Design the blueprint once, hire from it as many times as you like. More in Templates.

Option A — Hire from a template (fastest)#

Use this when a template already exists for the work you want done.

  1. Open Agents in the sidebar and select Hire New Agent. You'll see a three-step progress bar: Template → Configure → Deploy.
  2. Template — Choose a pre-built blueprint. Each card shows the template's name and a short description of what that agent does. Select one and choose Continue.
  3. Configure — Set the agent's identity:
    • Agent Name (required, 2–80 characters) — this is how the agent appears everywhere: Home, the team list, the inbox, activity, and approvals. If you leave it blank, the template's name is suggested.
    • Description (optional, up to 500 characters) — a short note on what this specific agent should focus on.
  4. Choose Continue to reach Deploy.
  5. Deploy — Review the summary (name, description, and the template it's based on). Choose Deploy Agent. To change anything, select Edit Configuration to step back.

That's it — the agent is created and started immediately. Jump to What deploy does.

Option B — Design an agent with the guided builder#

Use this when you want to shape the agent's role and behavior yourself. The builder asks a few questions and generates the agent's core files automatically. Open Templates in the sidebar and select New Template. It has three steps: Basics, Execution, and Guardrails.

Every long text field has a Write with AI button (it becomes Rewrite with AI once there's text). It drafts from what you've already filled in and lands as a suggestion you can Accept, Regenerate, or Discard — nothing is overwritten silently, and Accept keeps an Undo.

Step 1 — Basics (identity & objective)#

Field What it does
Template name (required) The agent's name and identity — at least 2 characters.
Description (optional) A one-line summary of the role, e.g. "Autonomous PM for engineering delivery and incident response."
Primary objective (required) The single outcome this agent exists to drive — at least 8 characters. Think in results: "Keep delivery predictable, unblock engineers, and reduce production risk." This is the north star the agent measures its decisions against.

Step 2 — Execution (responsibilities & behavior)#

Field What it does
Key responsibilities (required, one per line, up to 25) The concrete duties the agent owns — e.g. "Triage blockers from channels", "Track risky PRs and deploys", "Escalate approval-gated decisions." These become the work the agent looks for and does.
Communication style How chatty the agent is when it reports: Concise (short, to-the-point updates), Balanced (context when it matters), or Detailed (full reasoning included).
Heartbeat mode The agent's cadence — how it paces its check-ins: Adaptive (backs off when things are quiet) or Steady (a fixed check-in cadence). This is the rhythm of the agent's Observe phase.

Step 3 — Guardrails (notifications & memory)#

Field What it does
Owner name (optional) The person the agent reports to, so its messages read naturally.
Proactive notify on (one per line) Situations the agent should always surface — e.g. "incident detected", "deployment failure", "approval needed."
Do not notify on (one per line) Routine events to stay quiet about — e.g. "routine scans", "heartbeat completion" — so it doesn't create noise.
Integration hints (one per line, up to 30) The tools this role expects to use — e.g. "github", "slack", "kubernetes." These guide which integrations to connect.
Memory policy (optional) How long different memories should last — e.g. "Mappings weekly; decisions 90d; incident patterns until superseded." See Memory & learning.

Choose Create Template on the last step. Your new blueprint appears in the template library, ready to hire from using Option A above.

The builder's communication style, heartbeat, notify rules, and memory policy shape behavior. The rules about what an agent may do without asking — its supervision mode and the workspace guardrails — are governed separately. See Supervision & modes and Guardrails & sensitive actions.

What deploy does#

When you choose Deploy Agent, Syntrum:

  1. Creates the agent in your workspace with the name, description, and template you chose.
  2. Provisions the agent's own working environment and starts it immediately.
  3. Sends you to the Agents list, where your new hire appears among your team.

The "Starting up" state#

A brand-new agent's card may show a calm, animated Starting up state — "Spinning up [name]'s workspace — this only takes a moment." This means the agent's environment is still coming online. It is not an error and is deliberately shown differently from the red "needs attention" treatment. Once it's up, the card switches to a live runtime status.

Runtime status you'll see on the card:

Status Meaning
LIVE Running and healthy, ready for work.
BUSY Actively working a task right now.
IDLE Running but with nothing to do at the moment.
BLOCKED Waiting on you — usually an approval.
Starting up Workspace is still coming online (not an error).

You can Stop for now, Resume work, or Restart an agent from its card at any time. Stopping pauses in-progress work without deleting anything. Managing agents day to day is covered in Managing agents.

What's next#