Agent templates
An agent template is a ready-made agent you can turn into a real, editable draft in your project with one form. The catalog ships with the console (25 templates, versioned), so it works the same in the SaaS and on-premises editions and needs no network access. Templates are a starting point: the agent you get is an ordinary agent that you adapt like any other.
Open it from Build → Agents → Templates (also in the New agent menu and on the empty Agents page).
The gallery
The gallery lists every template as a card with its category, the capabilities it uses (knowledge, web search, client tools, approvals, sub-agents, skills, structured output, variables, schedule, webhook, sandbox, guardrails) and an availability badge. Filter by category or capability, or search by name. Featured templates (marked with * in the table below) are listed first. Selecting a card opens a detail drawer with the description, use cases, the instructions, tools and triggers the template brings, and a suggested first message to try.
Availability
Availability is read from your project, so a template never promises something the project cannot give:
| State | Meaning |
|---|---|
| Ready | The agent can be created and used as it is. Optional parts that are missing (a knowledge base, a web search provider) show as warnings. |
| Needs setup | The project lacks something the template relies on: a chat model, or a required knowledge engine or web search provider. The form explains what to add. An agent can still be created without a knowledge base or web search provider, but it cannot do its job until you add one. |
| Unavailable | The template needs the sandbox and this edition or license does not include it. |
What "Use template" creates
The Use template form is a full-screen form. You pick:
- Name and key (optional). The key defaults to the template key; if it is taken,
-2,-3and so on are added. A key you type yourself is refused with409 agent_key_conflictwhen it exists. - Model: any chat model of the project. The template only suggests a class (fast, balanced or reasoning).
- Knowledge bases for templates that use them. The agent is created without one if you pick none, and says so in the setup notes.
- Guardrails, recommended for templates that handle personal data.
- Sandbox template, for templates that need the sandbox (the default is
multi-basewhen your tenant has it). - Create the suggested triggers, and the timezone their schedules use (your browser's timezone by default).
- Publish version 1 right away. Off by default.
Creating is all-or-nothing. The console creates:
- A draft agent, with the instructions, tools, variables and structured output of the template, your model and picks, and a small mark of which template and version it came from.
- The template's skills, if it has any, under keys that start with
tpl-(for exampletpl-support-tone). A skill that already exists under that key is reused as it is, so edits you made to it are kept; a missing one is created. - The template's suggested triggers, if you left that on. They are always created disabled (a schedule does not run and a webhook does not accept deliveries until you enable them). Schedules use the timezone you chose.
- For a webhook trigger, the URL and the signing secret. The secret is shown once, on the result screen; afterwards only its last four characters are visible (you can rotate it on the trigger).
If any step fails, the agent, triggers and skills this call created are removed again and nothing is left behind. Publishing is the last step, so a failed publish also leaves nothing behind.
The result screen lists what is left to do: bind a knowledge base, add a web search provider, enable the triggers, publish, or connect the client tools (see below).
Client tools
Some templates (Customer Support, IT Helpdesk, Refund Approval) declare client tools with an approval. The agent asks your application to run the tool (such as create_ticket) and waits for the result. Such an agent works where a session can pause: the API or SDK with a session, an embedded chat or a portal. In the Studio playground you answer the call yourself. Connect the tool in your app before going live.
Webhook-triggered templates
Webhook templates (PR Reviewer, CI Failure Triage, Support Ticket Triage, Lead Qualifier, Email Triage) read the event from the message the trigger builds. Values taken from a webhook delivery reach the model only in the message, marked as external data; they are not placed in the system prompt.
The catalog
* marks a featured template.
| Template | Key | Category | Capabilities | Needs | Suggested triggers |
|---|---|---|---|---|---|
| Customer Support Agent * | customer-support | Support | Knowledge, Ask the user, Client tools, Approval, Skills, Variables, Guardrails | knowledge (optional), guardrails (recommended) | - |
| Support Ticket Triage | support-triage | Support | Structured output, Variables, Webhook | - | webhook (generic) |
| IT Helpdesk Assistant * | it-helpdesk | Support | Knowledge, Ask the user, Client tools, Approval, Variables | knowledge (optional) | - |
| Refund Approval Assistant | refund-approval | Support | Client tools, Approval, Variables | - | - |
| Knowledge Q&A * | knowledge-qa | Knowledge | Knowledge, Variables | knowledge (required) | - |
| HR Policy Assistant | hr-policy-assistant | Knowledge | Knowledge, Guardrails, Variables | knowledge (required), guardrails (recommended) | - |
| Pull Request Reviewer * | pr-reviewer | Engineering | Webhook, Structured output, Skills, Variables | - | webhook (github) |
| CI Failure Triage | ci-failure-triage | Engineering | Webhook, Structured output | - | webhook (github), webhook (generic) |
| Release Notes Writer | release-notes-writer | Engineering | Variables | - | - |
| Incident Postmortem | incident-postmortem | Engineering | Variables | - | - |
| SQL Assistant | sql-assistant | Engineering | Variables | - | - |
| Deep Research * | deep-research | Research | Web search, Sub-agents | web search (required) | - |
| Competitor Monitor | competitor-monitor | Research | Web search, Variables, Schedule | web search (required) | schedule 0 9 * * 1 |
| Daily Briefing * | daily-briefing | Research | Web search, Variables, Schedule | web search (required) | schedule 0 8 * * 1-5 |
| Data Analyst | data-analyst | Research | Sandbox | sandbox (required) | - |
| Lead Qualifier * | lead-qualifier | Sales and marketing | Web search, Variables, Structured output, Webhook | web search (optional) | webhook (generic) |
| Meeting Prep | meeting-prep | Sales and marketing | Web search, Variables | web search (optional) | - |
| Content writer | content-writer | Sales and marketing | Variables | - | - |
| Content repurposer | content-repurposer | Sales and marketing | Structured output, Variables | - | - |
| Meeting notes * | meeting-notes | Operations | Structured output | - | - |
| Email triage | email-triage | Operations | Structured output, Webhook | - | webhook (generic) |
| Document extractor | document-extractor | Operations | Structured output, Variables | - | - |
| Contract reviewer | contract-reviewer | Operations | Structured output, Skills | - | - |
| Translator | translator | Operations | Variables | - | - |
| Support Router * | support-router | Multi-agent | Sub-agents, Knowledge | knowledge (optional) | - |
Categories: Support, Knowledge, Engineering, Research, Sales and marketing, Operations, Multi-agent. The Deep Research and Support Router templates create sub-agents inline in the same agent (one agent is created, not several).
Adapting the agent
The created agent is a normal draft:
- Open it in the Studio and edit the instructions, model, tools, variables or structured output. Nothing is locked, and the agent does not change when the catalog changes (a later catalog version does not update agents you already created).
- Edit or delete the
tpl-skills, or add your own. - Add knowledge, web search, MCP servers or more triggers from the agent's Build and Ship tabs.
- Test with the playground (the detail drawer suggests a first message), then publish. API calls, triggers and A2A run the published version.
- To reuse your tuned agent elsewhere, export it as a manifest (
GET /api/agents/{agentId}/export) and import it into another project.
API
The dashboard API (session authentication, in the active project) has three routes. They are part of the Studio and are not on the client API (/api/client/v1).
| Route | Permission | Purpose |
|---|---|---|
GET /api/agents/templates | agents read | The catalog with the availability of each template in the project |
GET /api/agents/templates/{templateKey} | agents read | One template in full: spec, skills, suggested triggers, first message |
POST /api/agents/templates/{templateKey}/instantiate | agents write | Create the agent. Body: modelKey (required), name, description, key, knowledgeKeys, guardrailKeys, sandboxTemplateKey, createTriggers (default true), timezone (IANA, default UTC), publish (default false) |
instantiate answers 201 with agent, template (key, version, catalogVersion), triggers (disabled; a webhook trigger carries webhook.url and webhook.secret once), setup (what is left to do) and warnings (the notes a Studio save would show). Errors use the usual { error, code } shape: template_not_found (404), invalid_model, knowledge_not_found, knowledge_not_supported, guardrail_not_found, invalid_timezone, invalid_key, invalid_template_config (400), agent_key_conflict and sandbox_unavailable (409).