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

# Agents

> Automate routine admin work with agents that run on a schedule or when something happens, in all or selected locations, and connect an agent to your CRM or other apps through an MCP server

Agents automate routine operations in your coworking space. Each agent starts on a schedule or when something happens in your network, and does its job on its own in Spacebring and in the apps you [connect to it](#connect-other-apps). One emails the customer whose booking was just canceled, another publishes the week's events every Monday morning, a third chases an invoice nobody has paid.

## Create an agent

1. As an owner or a super administrator, go to **Network settings > AI & Agents**.
2. In the **Agents** section, click **Create agent**.

![AI & Agents settings with the Create agent button on Spacebring's coworking space management platform](https://mintcdn.com/spacebring/rNZ_8aOcm98_kqeJ/images/ai-agents-agents-ai-and-agents-settings.png?fit=max\&auto=format\&n=rNZ_8aOcm98_kqeJ\&q=85\&s=76ec193d5790699f6dac6cf37e172698)

> [!TIP]
>
> Instead of filling in the form, you can ask Spacebring Agent to create an agent for you. Click **Agent** in the left-side panel and describe the agent you need, for example "Create an agent that thanks members when their booking ends".

3. Enter a **Name** and add a photo. Members see them on anything the agent publishes, such as a feed post, so choose something your team and members will recognize, such as "Front Desk" or "Billing Assistant".
4. In **Triggers**, select what starts the agent. Start typing to search the list.
   * One event, for one job — for example, **Contract signed**, to email the customer what happens next.
   * Several events, when the agent should do the same thing for each of them — for example, **Booking created** and **Booking updated**. The agent follows the same instructions every time one of them occurs.
   * **Schedule**, for work that repeats at a set time — for example, a daily check of unpaid invoices.
5. Choose where the agent works:
   * Keep **All locations** on for an agent that serves your whole network.
   * Turn **All locations** off and select one or more **Locations** for an agent that serves only some of them — for example, a "Front Desk" agent for one location.
6. If you selected **Schedule**, set when the agent runs:
   * **Repeat** — **Daily**, **Weekly** with the weekdays to run on, or **Monthly** with the day of the month, from 1 to 28.
   * **Time** and **Timezone** — the agent runs at this time in the timezone you select.

![New agent form with triggers, locations, and a schedule on Spacebring's coworking space management platform](https://mintcdn.com/spacebring/WDLYHBSw6l3WH8KC/images/ai-agents-agents-new-agent-form.png?fit=max\&auto=format\&n=WDLYHBSw6l3WH8KC\&q=85\&s=247ed9ff01cf9abfab94b63a5d66dfaa)

7. In **Instructions**, describe [what the agent should do](#instructions).
8. Under **Permissions**, turn on every area the agent may access.

![New agent form with the instructions filled in and the Permissions switches on Spacebring's coworking space management
platform](https://mintcdn.com/spacebring/ue2rgCdbzdU4IdXQ/images/ai-agents-agents-instructions-and-permissions.png?fit=max\&auto=format\&n=ue2rgCdbzdU4IdXQ\&q=85\&s=385715c363324ef2eab129e768cdd41b)

9. Turn on **Enabled**, then click **Create**. The agent starts working on its next trigger.

> [!TIP]
>
> To start from a ready example, click one of the template links under the **Agents** section: follow up a canceled booking, follow up a
> signed contract, confirm a paid order, or welcome a new member. The form opens filled in, and you can change anything before you create
> the agent.

### Triggers and agent examples

An agent runs each time one of its triggers occurs in the locations it works in, and gets the details of what happened — the booking, the invoice, the ticket.

| Area | Triggers | Example agent |
| - | - | - |
| Schedule | Schedule | Every Monday at 09:00, publishes a feed post with the week's events in each location; on the 1st of each month, emails an administrator a [report](/docs/guides/send-automatic-reports-and-digests.md) on last month's revenue and new members |
| Bookings | Booking created, Booking updated, Booking paid, Booking ended, Booking canceled | On **Booking canceled**, emails the customer three free slots for the same room that week |
| Check-ins and visits | Checked in, Checked out, Visit created, Visitor checked in, Visit request created, Visit request approved, Visit request rejected | On **Checked in**, emails a day-pass guest the Wi-Fi password and printing rules from your guides; on **Visit created**, emails your front-desk administrator the visitor's name, host, and arrival time |
| Customers | Customer created, Request to join created, Request to join approved | On **Request to join created**, emails an administrator the person's name, company, and the plan they asked about |
| Contracts | Contract issued, Contract signed, Contract declined, Contract terminated | On **Contract signed**, emails the customer the access details and house rules from your location guides |
| Subscriptions | Subscription purchased, Subscription started, Subscription renewed, Subscription ending soon (three days before a subscription's end date), Subscription expired | On **Subscription ending soon**, emails the member the plans they can continue on |
| Invoices | Invoice created, Invoice issued, Invoice paid, Invoice voided | On **Invoice issued**, asks customers with no saved payment method to add one so their next invoice is paid automatically |
| Events | Event ticket created, Event ticket paid | On **Event ticket paid**, emails the attendee the address, start time, and how to check in |
| Orders | Order created, Order paid, Order completed, Order canceled | On **Order completed**, emails the customer a thank-you and asks how the purchase worked out |
| Support | Support ticket created, Support ticket status updated, Support ticket comment created | On **Support ticket status updated**, emails the customer a summary once their ticket is solved |
| Chats | Chat message received (someone messages the agent in chats) | On **Chat message received**, answers a member's question about opening hours or a free meeting room in the same chat |

### Instructions

The instructions are the agent's whole brief: it sees which trigger started the run, the details that came with it, and the data in the areas you allowed. Write them the way you would brief a new team member:

* Say what to look up first, such as the customer, the resource, or your location guides.
* Say what to produce and who gets it: an email to the customer or to an administrator, a feed post, an updated item.
* Say when the agent should do nothing, for example: "If the booking was canceled by an administrator, do nothing."
* Say which language to write in, for example: "Write in the customer's language."
* Give one job to one agent. For a second job, create a second agent.

An agent can:

* Look up anything in your network, from customers and bookings to invoices, orders, events, visits, tickets, guides, sent emails, and settings.
* Email one customer of its locations or one administrator at a time, in your network's branding, with a CSV file attached when a list is too long for the email.
* Publish and edit feed posts.
* Message people in chats and create group chats.
* Open support tickets, comment on and reply to them, and change their status, assignee, or type.
* Invite members, add nonmembers, create companies, move members to a company or remove them from it, remove members and companies, and delete nonmembers.
* Create, copy, and edit resources, products, events, benefits, guides, plans, packages, discounts, currencies, and categories, and delete resources, plans, and coupons.
* Book resources for customers, and change or cancel bookings. An agent doesn't run again for a booking it created, changed, or canceled itself.
* Assign resources to customers for a period without a subscription, and change or remove those assignments.
* Change or cancel subscriptions and their items, and issue, re-date, terminate, or delete contracts.
* Create, issue, void, and delete invoices, change their items, and issue or void credit notes.
* Add or deduct a customer's credits and day passes.
* Change location settings.
* Use the apps you [connect to it](#connect-other-apps).
* Search the web and the Spacebring documentation. An agent started by an event can't open links or download files from them, so text that a member or visitor writes can't make it send your data elsewhere.

### Permissions

Without any area turned on, an agent can still look up most data in its locations, like an administrator without permissions. Turn on the areas where its job needs to make changes — each one covers the same ground as the [administrator permission](/docs/administration/permissions.md) of that name:

* **Customers** — customer profiles and their data.
* **Billing** — contracts, subscriptions, invoices, plans, and discounts.
* **Resources** — resources, bookings, and occupancy.
* **Feed** — posts and comments.
* **Events** — events and tickets.
* **Shop** — products, categories, and orders.
* **Benefits** — benefits, categories, and applications.
* **Support** — support tickets and guides.
* **Settings** — location settings and integrations.

The permissions apply only to the agent's locations.

> [!WARNING]
>
> An agent never asks for approval. A run has nobody to ask, so the agent carries out actions such as sending an email to a customer, in
> Spacebring and in the apps you connect to it. It works like a chat set to [Auto-approve](/docs/ai-agents/create-edit-configure.md#auto-approve-actions). Turn on only the areas the agent's task needs.

## Review runs

Every start of an agent is recorded as a run, and the AI usage of each run is charged to your [network balance](/docs/ai-agents/network-balance.md). Open an agent on the **Network settings > AI & Agents** page to see its **Runs**, newest first.

* A run shows the trigger that started it, the date, and what the agent said when it finished.
* **Running** means the agent is still working, **Completed** that it finished, and **Failed** that it stopped with an error, which the run shows. A failed run isn't started again, so the agent never repeats an email or a change it already made.
* **Skipped** means the agent didn't run because your network's balance ran out. Network owners are notified, and runs resume once the balance is [topped up](/docs/ai-agents/network-balance.md#top-up-the-balance).
* Click a run to open the instructions the agent followed, every change it made, the CSV files it wrote, and its full reply. Each change shows what the agent did and to whom — for example, the subject of an email and the address it went to — and a failed run still lists the changes made before it stopped.

![An agent's list of runs on Spacebring's coworking space management platform](https://mintcdn.com/spacebring/WDLYHBSw6l3WH8KC/images/ai-agents-agents-agent-runs.png?fit=max\&auto=format\&n=WDLYHBSw6l3WH8KC\&q=85\&s=3f8649b02f92d71035c51c7989aac8eb)

> [!TIP]
>
> Agents can make mistakes. Test a new agent before you rely on it: cause its trigger yourself a few times, or set its schedule to the next
> quarter of an hour, then read those runs and correct the instructions where the agent went the wrong way.

## Manage agents

On the **Network settings > AI & Agents** page, click the **(...)** menu next to an agent:

* **Disable** stops the agent from running and keeps it in the list with its past runs. **Enable** starts it again.
* **Edit** changes the name, photo, triggers, locations, schedule, instructions, and permissions. The agent's next run follows the new setup.
* **Delete** removes the agent for good: it stops running and disappears from the list, and this cannot be undone.

## How agents appear in your network

An agent is a separate user in your network, under the name and photo you gave it.

* A feed post it publishes carries that name and photo, like a post from any other author.
* A chat message it sends comes from that name and photo too.
* The emails it sends go out under your network's name, like every other Spacebring email.

![A feed post published by an agent under its name and photo on Spacebring's coworking space management platform](https://mintcdn.com/spacebring/WDLYHBSw6l3WH8KC/images/ai-agents-agents-feed-post-by-an-agent.png?fit=max\&auto=format\&n=WDLYHBSw6l3WH8KC\&q=85\&s=3627be2158a816ade8d7829b13c21d05)

## Connect other apps

Connect another app, such as your CRM, so an agent can work in it: look up a deal, read a page, or create a task. An app connects through its MCP server. The app's help center gives the server's URL and how to sign in.

### Connect an app

1. Open the agent on the **Network settings > AI & Agents** page.
2. In the **MCP servers** section, click **Connect MCP server**.

![Agent page with the MCP servers section and the Connect MCP server button on Spacebring's coworking space management
platform](https://mintcdn.com/spacebring/OZdWT00aG4z3g_Fo/images/ai-agents-agents-connect-an-app-1.png?fit=max\&auto=format\&n=OZdWT00aG4z3g_Fo\&q=85\&s=882b73446f2917875c903d5304ebe0d6)

3. Enter a **Name**, such as "Notion".
4. Enter the **URL** of the app's MCP server, such as `https://mcp.notion.com/mcp`.
5. Optionally, in **Description**, say what the app is for and when the agent should use it, such as "Our CRM. Use it for leads and deals."
6. In **Authentication**, choose how Spacebring gets access to the app:
   * **OAuth sign-in** — you sign in to the app on its website. Fill in **Client ID** and **Client secret** only if the app asks you to register Spacebring yourself.
   * **API key** — enter a key you created in the app.
   * **API key in a custom header** — enter the **Header name** from the app's help center, such as `X-API-Key`, and the **API key**.
   * **No authentication** — for an app that needs no sign-in.

![Connect MCP server form with a name, URL, and description filled in and OAuth sign-in selected on Spacebring's coworking space
management platform](https://mintcdn.com/spacebring/OZdWT00aG4z3g_Fo/images/ai-agents-agents-connect-an-app-2.png?fit=max\&auto=format\&n=OZdWT00aG4z3g_Fo\&q=85\&s=ef79962d291d789073e5c7a792804878)

7. Click **Connect**. For **OAuth sign-in**, sign in to the app and allow access, and you come back to Spacebring.

The agent uses the app from its next run, as the account you signed in with or with your API key. An app is connected to one agent only, so connect it to each agent that needs it.

> [!TIP]
>
> You can also connect an app from a chat. Open the agent's page, click **Ask Spacebring** in the top right corner, and ask, for example, "Connect Notion to this agent".

### Manage connected apps

In the agent's **MCP servers** section, click the **(...)** menu next to an app:

* **Disable** stops the agent from using the app. **Enable** turns it back on.
* **Edit** changes the app's name, URL, description, and API key. For an app connected with **OAuth sign-in**, you can change only the name and description.
* **Delete** removes the app with its sign-in or API key, and this cannot be undone.

The agent can't use an app marked **Waiting for sign-in** or **Sign-in expired**. To sign in again, delete the app and connect it again.


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