> ## Documentation Index
> Fetch the complete documentation index at: https://help.clinikehr.com/llms.txt
> Use this file to discover all available pages before exploring further.

# AI agents

> Build an agent on the canvas, choose what starts it, and deploy it — plus what your plan lets you publish and what each step costs.

export const StillStuck = ({topic}) => <div className="ck-callout ck-callout--note" role="note">
    <p className="ck-callout__body">
      <strong>Still stuck{topic ? ` with ${topic}` : ''}?</strong>{' '}
      Check <a href="/platform/help/troubleshooting">Troubleshooting</a> first —
      it covers the failures we see most often. If that does not resolve it,{' '}
      <a href="/platform/help/contact-support">contact support</a> and include
      your workspace name and the time the problem happened, so we can find the
      matching entry in your audit log.
    </p>
  </div>;

export const TaskHeader = ({before, time, after}) => <div className="ck-task">
    {before ? <div className="ck-task__cell">
        <div className="ck-task__title">Before you start</div>
        <p className="ck-task__body">{before}</p>
      </div> : null}

    {time ? <div className="ck-task__cell">
        <div className="ck-task__title">Takes about</div>
        <p className="ck-task__body">{time}</p>
      </div> : null}

    {after ? <div className="ck-task__cell">
        <div className="ck-task__title">When you're done</div>
        <p className="ck-task__body">{after}</p>
      </div> : null}
  </div>;

export const Path = ({steps = []}) => <span className="ck-path">
    {steps.map((s, i) => <span key={i} className="ck-path__step">
        {i > 0 ? <span className="ck-path__sep" aria-hidden="true">
            ›
          </span> : null}
        {s}
      </span>)}
  </span>;

export const Availability = ({editions = ['all'], plans, roles, note}) => {
  const list = editions.includes('all') ? ['clinic', 'pharmacy', 'lims'] : editions;
  return <div className="ck-avail" role="note" aria-label="Feature availability">
      <span className="ck-avail__label">Available in</span>

      {list.map(e => <span key={e} className={`ck-pill ck-pill--${e}`}>
          {EDITION_LABELS[e] || e}
        </span>)}

      {plans ? <span className="ck-avail__label">Plan</span> : null}
      {plans ? <span className="ck-pill ck-pill--plan">{plans}</span> : null}

      {roles ? <span className="ck-avail__label">Who</span> : null}
      {roles ? <span className="ck-pill ck-pill--role">{roles}</span> : null}

      {note ? <span className="ck-avail__note">{note}</span> : null}
    </div>;
};

<Availability editions={['clinic']} plans="Starter, Essential and Team — the builder is closed on Free" roles="Owners and managers" note="Agent and Classify steps consume credits" />

An agent is a small automation you draw rather than code: something starts it, it does a few steps, it stops. Use it for jobs nobody wants to do by hand — triaging an inbox, drafting a follow-up, tagging a form.

<Path steps={['Sidebar', 'AI Agents']} />

<TaskHeader before="A paid plan, and credits on your account before anything runs" time="Fifteen minutes for a first working agent" after="A deployed agent that runs when its trigger fires" />

## What you're looking at

The page is headed **Agent Builder** — "Build AI agents and workflows with custom logic and tools." The header carries **Runs** (every execution, with what it did and what it cost), **+ Create** — "Name your agent. You can design it on the canvas next." — and **Public UIs** once your plan includes any. Below sits **Your agents** and a template gallery.

The node palette is grouped:

| Group     | Nodes                                       |
| --------- | ------------------------------------------- |
| **Core**  | **Agent**, **Classify**, **End**, **Note**  |
| **Tools** | **File search**, **Guardrails**, **MCP**    |
| **Logic** | **If / else**, **While**, **User approval** |
| **Data**  | **Transform**, **Set state**                |

Every graph also has a **Start** node, added for you. Only **Agent** and **Classify** consume credits — they carry a **billable** pill in the inspector. Everything else is free to run: **File search** is "Deterministic — no model, no credit."

Across the top of the canvas sit the status chips — "● Live · v3" for what is deployed and "Draft · v4" for what you are editing — plus a **Saved** indicator and your credit balance.

On the **Free plan** there is no canvas at all: the page shows an upgrade panel for **Agent Studio** — "Build AI agent virtual assistants on any paid plan."

## Build and deploy an agent

<Steps>
  <Step title="Create the workflow">
    Select **+ Create**, name it — "Agent name (e.g. Appointment reminder)" — and select **Create**. The canvas opens.
  </Step>

  <Step title="Choose what starts it">
    Select the **Start** node and set its **Trigger**: **Manual / Preview** for testing; a **Schedule** of **Every 15 minutes**, **Every 30 minutes**, **Every hour**, **Every day** or **Every week**; **Webhook / API call**; an inbound message (**Inbound SMS**, **WhatsApp**, **email**, **fax** or **phone call (voice)**); or one of around fifty events such as **Appointment booked**.
  </Step>

  <Step title="Lay out the steps">
    Drag nodes from the palette and connect them. Start with one **Agent** node and an **End**; add **Classify** and **If / else** once that works. Put a **User approval** node in front of anything that sends, charges or writes to a record.
  </Step>

  <Step title="Test it by hand">
    Run it manually and read the result under **Runs** — this is where you find out what your instructions actually said, as opposed to what you meant.
  </Step>

  <Step title="Deploy it">
    Select **Deploy**. A toast confirms it — "Deployed — v1 is now live" — and the chip changes to **Live**. Later edits sit in the draft until you **Redeploy**.
  </Step>
</Steps>

<Note>
  **A saved draft does nothing.** The builder says so under the trigger: "Event & schedule triggers take effect after you Deploy." A **Webhook / API call** trigger needs one more step — the endpoint has to be switched on under **Settings** → **AI settings**, and that card is a **Team** feature. See [AI settings](/practice/settings/ai-settings).
</Note>

<Warning>
  **Agent** and **Classify** steps consume credits every time they run, and a schedule multiplies that: **Every 15 minutes** is 96 runs a day, roughly 2,900 a month, times the billable nodes in the flow. Start on **Every day**, watch the cost under **Runs**, and shorten the interval only once you know what one run costs. See [Agent credits](/platform/ai/credits).
</Warning>

## What your plan allows

Drafts are unlimited on every paid plan; publishing is what is capped.

| Plan      | The builder                        | Deployed agents | Public agent UIs                                      |
| --------- | ---------------------------------- | --------------- | ----------------------------------------------------- |
| Free      | Not available — upgrade panel only | 1               | 0                                                     |
| Starter   | Yes                                | 5               | 2                                                     |
| Essential | Yes                                | Unlimited       | 5                                                     |
| Team      | Yes                                | Unlimited       | Unlimited, plus a custom link and the webhook trigger |

The limit is reported when you try to deploy past it: "Your Starter plan can publish up to 5 agents or automations at a time. Drafts are always unlimited. Upgrade to Essential to publish more." Undeploying one frees the slot, so you can rotate rather than upgrade.

## Where the depth lives

This page is the map; the detail sits in the platform section:

* [Agents](/platform/ai/agents) — writing an instruction that behaves.
* [Agent tools](/platform/ai/agent-tools) — what **File search** and **MCP** reach.
* [Publishing agents](/platform/ai/publishing-agents) — public UIs and webhooks.
* [Agent credits](/platform/ai/credits) — what a step costs.
* [AI safety](/platform/ai/safety) — guardrails and approvals.

## Who can do this

| Action                               | Who by default | The permission, in the product's words                                   |
| ------------------------------------ | -------------- | ------------------------------------------------------------------------ |
| Open **AI Agents** and draft         | Owner, manager | Role — no permission of its own                                          |
| Deploy or undeploy an agent          | Owner, manager | Role                                                                     |
| Publish a public agent UI            | Owner, manager | Role                                                                     |
| Configure the webhook trigger        | Owner, manager | Role — "Only clinic owners and managers can configure webhook triggers." |
| See the credit balance and run costs | Owner, manager | **View financial dashboard**                                             |

* **Enforcement is opt-in, per person.** Someone never saved in the permissions sheet is unrestricted, and the sheet warns **"Saving starts enforcing"**.
* **A denied action is recorded; an allowed one is not.** Hiding **AI Agents** from a sidebar is a convenience, not a security boundary. A credit figure you may not see renders as a dash, not a zero.

See [Staff permissions](/platform/team/permissions).

## Check it worked

* The chip at the top of the canvas reads **Live** with a version number, not just **Draft**.
* A manual run appears under **Runs** with its steps and its cost.
* A scheduled agent produces a run at the next interval without you touching anything.
* A webhook agent produces a run when the other system calls it — nothing at all means the endpoint is still disabled in **AI settings**.

## Common issues

<AccordionGroup>
  <Accordion title="My scheduled agent never runs">
    It is still a draft. "Event & schedule triggers take effect after you Deploy" — select **Deploy**.
  </Accordion>

  <Accordion title="My webhook agent does not fire">
    Deploying is only half of it. The endpoint has to be enabled under **Settings** → **AI settings**, and that card is a Team feature. See [AI settings](/practice/settings/ai-settings).
  </Accordion>

  <Accordion title="Deploy failed and mentions my plan">
    You are at your plan's deployed limit: "Your Starter plan can publish up to 5 agents or automations at a time. Drafts are always unlimited. Upgrade to Essential to publish more." Undeploy one, or move up a plan.
  </Accordion>

  <Accordion title="Credits are draining faster than expected">
    Check the schedule interval and the number of **Agent** and **Classify** nodes — every one bills on every run. **Runs** shows the cost per execution.
  </Accordion>

  <Accordion title="The agent did something I did not want sent">
    Put a **User approval** node in front of the sending step and redeploy. See [AI safety](/platform/ai/safety).
  </Accordion>

  <Accordion title="There is no Public UIs button">
    Your plan includes none — "Public Agent UIs are available on the Starter plan and above." See [Publishing agents](/platform/ai/publishing-agents).
  </Accordion>
</AccordionGroup>

## FAQ

<AccordionGroup>
  <Accordion title="Does drafting cost anything?">
    No. Drafts are unlimited on every paid plan, and credits are consumed only when an **Agent** or **Classify** step actually executes.
  </Accordion>

  <Accordion title="Can an agent see my clients' clinical records?">
    Only through the tools you give it, and within the safeguards in [AI safety](/platform/ai/safety). It has no standing access to anything you have not wired to it.
  </Accordion>

  <Accordion title="What happens to a live agent if I edit it?">
    Your edits sit in the draft — the chips show **Live · v3** and **Draft · v4** side by side. The version running is the one you last deployed.
  </Accordion>

  <Accordion title="Can I stop an agent without deleting it?">
    Yes. Undeploy it — the design is kept and the slot is freed.
  </Accordion>

  <Accordion title="What is the difference between Agent and Classify?">
    **Classify** sorts an input into one of your categories and pairs with **If / else** to branch. **Agent** reasons and writes. Both bill, but **Classify** is the cheaper way to route.
  </Accordion>
</AccordionGroup>

<StillStuck topic="AI agents" />