Quickstart

Build your first AI agent in 10 minutes

You'll create a Support Triage Agent that reads a ticket, classifies it, drafts a response, and logs the result. By the end you'll know the full agent loop end-to-end.

What you'll learn
  • How to create an agent with the 9-step builder
  • How to attach an LLM model and a tool
  • How to apply a guardrail profile
  • How to test the agent and read the execution trace

Before you start

Sign in to your workspace, then click Agents in the sidebar. You should see your agent library — empty if this is your first agent. Click + Create Agent to start the builder.

Create the agent

  1. 1

    Name it

    Call it "Support Triage Agent". Add a one-sentence description. Pick "Support" as the category.
  2. 2

    Pick an LLM

    Choose a fast, low-cost model — GPT-4o-mini, Claude Haiku, or your provider's equivalent. You can change this later.
  3. 3

    Attach a tool

    Add the Zendesk or Jira integration. The agent will use it to read tickets and post updates. If you haven't connected one yet, follow the OAuth flow.
  4. 4

    Skip skills, knowledge, memory, voice

    Defaults are fine for the first agent. We'll layer these in later guides.
  5. 5

    Add a guardrail profile

    Pick the Standard preset. It blocks abusive language, jailbreaks, and PII leakage out of the box.
  6. 6

    Review and publish

    Confirm the summary and click Publish. The agent is now production-ready inside your workspace.

Test it

  1. 1

    Open the test view

    Click Test on the agent card. A chat panel opens with full execution trace below each response.
  2. 2

    Send a real ticket

    Paste a sample ticket: "Customer says their invoice is wrong, urgent, mentions refund." Send.
  3. 3

    Inspect the trace

    Expand the response. You'll see the LLM call, the tool invocation (ticket read or update), token usage, latency and cost. This is what every production run records.

Frequently asked questions

How much does running an agent cost?
Cost is the LLM token spend plus integration calls. The execution trace shows a per-run dollar amount so you can size pilots accurately.
Can I edit the agent after publishing?
Yes. Click Edit to re-enter the builder. Changes are versioned — you can roll back if a new version regresses.
Where does the agent run?
Inside your Dezifi workspace. Each tenant is isolated. On-premise and private-cloud deployments run inside your own infrastructure.