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

# Internal agents

> What internal agents are, how they differ from customer agents and workflows, and how a run works from trigger to result.

Internal agents are AI agents that work for your team instead of talking to your customers by default. They run operational work behind the scenes: checking arrivals, triaging tasks, summarizing escalations, reacting to PMS events, answering teammates in Slack, or calling your other systems.

An internal agent doesn't own an inbox. It starts when one of its triggers fires, reasons about the job you gave it, uses the tools you enabled, and then finishes, posts a result, or hands work to another agent.

## Internal agents, customer agents and workflows

| | Customer agent | Internal agent | Workflow |
| - | - | - | - |
| **Works for** | Your contacts and customers | Your team and your systems | Your team and your systems |
| **Starts from** | Messages arriving in its inbox | Messages, a schedule, a webhook, a platform event, or a teammate | A workflow trigger |
| **Decides how** | AI reasoning | AI reasoning | Fixed steps you build |
| **Replies to contacts** | Yes, in its inbox | Only where you set **Reply**, or through a tool | Only through a send step |
| **Sandbox testing** | Yes | No. Every run is real. | Not applicable |
| **Credits** | Doesn't consume credits | Consumes credits | Paid actions consume credits |

Use a workflow when something should happen the same way every time. Use an internal agent when the right action depends on judgment, or when there are too many cases to list in advance. For example, a workflow can react to "cleaning status changed", but deciding which guest to tell, and whether to tell them at all, is a job for an agent.

## What an internal agent can do

* **Read your workspace.** Look up reservations, contacts, conversations, tasks, listings and Knowledge through the [tools](https://docs.conduit.ai/build-your-agent/tools.md) you enable.
* **Act.** Send messages, update records, and call connected integrations, MCP servers, custom tools and AI-triggered workflows, depending on the tools you enable.
* **Follow your process.** Use [Skills](https://docs.conduit.ai/build-your-agent/skills.md) for repeatable procedures and [guardrails](https://docs.conduit.ai/build-your-agent/agents/persona-and-guardrails.md) for rules it must always follow.
* **Delegate.** Hand parts of a job to [subagents](https://docs.conduit.ai/build-your-agent/agents/internal-agents/subagents.md) that run in parallel.
* **Talk with your team.** Answer in a chat in Conduit, or in Slack and Microsoft Teams when you turn those on.

## What starts a run

The first four triggers each have a page under **Runs on** in the agent's sidebar. Teammates can also start the agent directly.

| Trigger | Runs when | Learn more |
| - | - | - |
| **Messages** | A message arrives in Slack, Microsoft Teams, a phone number, an email address, a WhatsApp number or a website widget you choose. | [Messages](https://docs.conduit.ai/build-your-agent/agents/internal-agents/messages.md) |
| **Schedule** | A time you set comes around, such as every weekday at 9 AM. | [Schedules](https://docs.conduit.ai/build-your-agent/agents/internal-agents/schedules.md) |
| **Webhook** | An external system sends JSON to the agent's private endpoint. | [Webhooks](https://docs.conduit.ai/build-your-agent/agents/internal-agents/webhooks.md) |
| **Platform events** (the **Workflow** row) | A platform event happens, such as a new reservation, a task being assigned, or a set time before check-in. | [Platform events](https://docs.conduit.ai/build-your-agent/agents/internal-agents/platform-events/overview.md) |
| **Chat** | A teammate starts a chat with the agent, or Operator hands work to it. | [Chat and activity](https://docs.conduit.ai/build-your-agent/agents/internal-agents/chat-and-activity.md) |

An agent can have several triggers at once. Every run, whatever started it, appears in the agent's **Activity** tab.

## How a run works

**1. A trigger fires**

Conduit starts a run and passes the agent what it needs: the message, the event payload, the webhook body, or the task you wrote for the schedule.

**2. The agent plans**

The agent reads its instructions, guardrails and Skills, then decides what to do with this input.

**3. The agent uses tools**

The agent looks things up and takes actions with the tools enabled on its **Tools** tab. It can delegate parts of the job to subagents.

**4. The run ends**

The agent finishes its work. In a chat, Slack or Teams it answers the person who asked. On a message trigger set to **Reply**, Conduit sends its answer back in the same conversation.

When AI is paused on a conversation or contact, message triggers set to **Reply** don't run there. **Listen only** agents keep running. Message triggers also skip messages Conduit marks as spam.

## Agent status

Each internal agent is either **Active** or **Inactive**. Set it under **Agent status** in the status menu in the agent's top bar, then select **Save**. See [Turn the agent on or off](https://docs.conduit.ai/build-your-agent/agents/internal-agents/create-and-configure.md#turn-the-agent-on-or-off).

| Status | What happens |
| - | - |
| **Active** | The agent runs from every enabled trigger. |
| **Inactive** | The agent is turned off. Triggers don't start it, and its webhook endpoint rejects requests. **Run now** can still start a real run. |

Turning a single trigger off, such as one schedule or the webhook, leaves the agent's other triggers running.

## Credits

Internal agent runs consume credits. Each run has a dynamic cost based on the work the agent does, so there is no flat rate per run. Chats with an internal agent are runs too, and they also consume credits.

To see what each internal agent has used, go to **Settings > Billing** and expand **Internal Agents** for a per-agent breakdown. Open an agent to see its credits and runs for the current billing cycle. See [Credits](https://docs.conduit.ai/admin/credits.md) for how credits work across Conduit.

## Next steps

> [card] [Create and configure](https://docs.conduit.ai/build-your-agent/agents/internal-agents/create-and-configure.md)
>
> Create an internal agent and set its instructions, tools, Skills and settings.

> [card] [Chat and activity](https://docs.conduit.ai/build-your-agent/agents/internal-agents/chat-and-activity.md)
>
> Talk to the agent, run it on demand, and review what it did.

> [card] [Use cases](https://docs.conduit.ai/build-your-agent/agents/internal-agents/use-cases.md)
>
> Internal agents other teams have built.

> [card] [Subagents](https://docs.conduit.ai/build-your-agent/agents/internal-agents/subagents.md)
>
> Split a large job across specialized agents.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.