SyntrumDocs
Open app

Templates

A template is a reusable blueprint for an agent: its identity, objective, responsibilities, communication style, and operating guardrails, captured once so you can design once and deploy many. When you hire from a template, the agent starts pre-configured instead of blank.

Open Templates from the sidebar (under Building Agents).

The template library#

The library is a grid of cards. The first card is always Blank Template — "Build your own agent behavior from scratch," where you define everything. The rest are your saved templates plus curated presets.

Each template card shows:

  • Name and description of the blueprint.
  • A scope indicator describing how independently agents from this template run — for example Starter (simple rules), Smart (decision making), or Autonomous (acts without approval). Custom templates read Custom — You define everything.
  • Capability badges — the kinds of tools and abilities the template expects (Email, CRM, Calendar, Decision making, Human approval, and so on).
  • A category and a rough count of agents deployed from it.

Two actions per card:

  • Create Agent — hire a new agent from this template. This opens guided creation pre-filled from the template.
  • Edit Template — open the template to change it.

At the top of the page, Create Agent from Template and New Template give you the same two paths from the header.

Templates are workspace-level. A blueprint you save is available to everyone who can hire agents, so your team stays consistent — every "Sales Follow-up" agent behaves the same way out of the box.

What a template captures#

Guided creation walks you through three steps. Everything you enter is saved into the template and used to generate the agent's core configuration files automatically.

Step 1 — Basics (identity & objective)#

  • Template name (required, at least 2 characters) — e.g. "Allan K - Autonomous TPM."
  • Description (optional) — a short summary of what this blueprint is for.
  • Primary objective (required, at least 8 characters) — the single outcome the agent exists to drive, e.g. "Keep delivery predictable, unblock engineers, and reduce production risk."

Step 2 — Execution (responsibilities & behavior)#

  • Key responsibilities — one per line, up to 25. These are the concrete jobs the agent owns, e.g. "Triage blockers from channels," "Track risky PRs and deploys," "Escalate approval-gated decisions."
  • Communication style — Concise (short, to-the-point), Balanced (context when it matters), or Detailed (full reasoning included).
  • Heartbeat mode — Adaptive (backs off when quiet) or Steady (fixed check-in cadence). This governs how often the agent wakes up to check on its world.

Step 3 — Guardrails (notifications & memory)#

  • Owner name (optional) — who the agent reports to.
  • Proactive notify on — situations the agent should proactively flag you about (one per line), e.g. "incident detected," "deployment failure," "approval needed."
  • Do not notify on — routine noise to stay quiet about, e.g. "routine scans," "heartbeat completion."
  • Integration hints — the tools this agent expects to use (one per line, up to 30, 80 characters each), e.g. github, slack, kubernetes. These are hints for which systems to connect — actual access is granted later in Connecting tools.
  • Memory policy (optional) — how long different kinds of memory should persist, e.g. "Mappings weekly; decisions 90d; incident patterns until superseded."

Write with AI. Most text fields have an AI assist button in the corner. It drafts from the context you've already filled in, or rewrites what's there. Suggestions never overwrite silently — you get Use this / Replace text, Regenerate, and Discard, and Accept keeps an Undo.

The stepper#

The three steps — Basics, Execution, Guardrails — appear as a numbered stepper at the top. You can click backward freely; moving forward validates the current step first (a valid name and objective before leaving Basics, at least one responsibility before leaving Execution). Previous and Next move you through, and the final step's button is Create Template. On save you return to the library with the new template ready to deploy.

Reusing a template#

To hire from a template, click Create Agent on its card (or Create Agent from Template in the header). Guided agent creation opens pre-filled with the blueprint's fields, so you only adjust what's specific to this hire — the name, any tweaks to responsibilities — and deploy. The new agent goes through the Starting up state and then joins My Team.

Creating and editing templates, and hiring agents from them, requires an Admin or Owner role. See Team & roles.

What's next#