AI Agent

Ask the AI Agent to find information, draft a reply, or carry out permitted work. Include the property, reservation, conversation, or task involved. Configure its knowledge and permissions in AI Hub.

​

Use the AI chat sidebar

Open the AI sidebar from the page you are working on. Type @ to attach the relevant property, person, file, conversation, reservation, task, template, agent, or workflow. Then describe the result you want: “Summarize this guest request and create a maintenance task for tomorrow,” for example. For a task, include the department, priority, timing, and assignee if known. Answer clarifying questions in the same chat. If an action requires approval, review the exact target and proposed change before deciding. Open the resulting task to confirm it was created with the requested details.

​

Review suggested replies in Inbox

With Autopilot off, the agent can prepare a suggested reply for review. Read it alongside the latest message, selected reservation, and reply channel. Edit or regenerate it when needed, then choose whether to send it. A suggested reply has not already gone to the guest. Open AI Trace or the available sources to inspect the records and tool results used for the reply. Check the source when the answer makes a specific claim about a time, fee, policy, or property feature. Use Teach AI when the underlying information needs correction.

​

Handle an escalation

Read the escalation reason and any acknowledgement already sent. Assign a person to resolve the request, then continue from the conversation’s current state so the guest does not receive conflicting answers. An escalation notification needs a configured destination; it is not enough to mention a team in the agent’s instructions.

​

Write clear agent instructions

Set the agent’s role and tone in AI Hub. Put scenario-specific handling in Policies and property facts on the property.

​

Control permissions

In Access & Safety, set each tool to Allowed, Ask first, or Off. Ask first requires review even when the autonomy mode otherwise allows actions within limits. See Agent permissions and approvals.

​

Look up operating details

Ask about the record and time period you need. For example, “Show open maintenance tasks for Apartment 4 tomorrow” is easier to verify than “What needs doing?” The agent can look up permitted reservations, properties, tasks, reviews, contacts, and available connected calendar information. If several records match, resolve the ambiguity before a change. Check the property, guest, and dates in any linked result.

​

Use documents, skills, and memory

AI Hub controls the agent’s approved knowledge.

• Documents contain reference material. Use Policies for scoped handling rules and property records for facts that vary by home.
• Skills are step-by-step instructions the agent can follow.
• Memory helps the agent remember approved preferences and repeat procedures.
• Permissions decide which agents can use which documents and skills.

Trellis keeps conversation history separate from approved memory. The agent can search earlier sessions when it needs the original context, while durable memory stores only selected preferences, corrections, and procedures. Workspace-wide memory is available across the agent’s authorized work; property memory is added only when the agent and teammate can access that property. Workspace admins can review an agent’s memories in AI Hub. From there, they can search memory, add or correct an item, archive and restore it, or permanently forget it for privacy. Archiving removes an item from future answers without deleting its lifecycle history. Use separate agents when different jobs need different rules. For example, one agent may handle guest replies while another supports owner updates or team operations.

​

Read Slack files and shared context

When Slack is connected and permitted, the agent can use supported channel messages and files. It can summarize shared text files, describe supported images, and use that context while answering team questions. The agent needs a connected Slack account with access to the channel and files.

​

Review linked records in chat

When the agent references a record, Trellis can show record pills, inline lists, and detail panels. Use these to check the source before you act. You may see:

• A reservation, property, task, contact, review, or workflow link.
• An inline list, such as several matching tasks or reservations.
• A detail panel with the related record and a breadcrumb trail when you click deeper.

​

Review agent activity

Use Insights > Pulse to review recent AI activity. Pulse helps managers see what the agent did, where it happened, and whether anything needs attention. Activity details can show:

• The conversation, task, property, or reservation involved.
• The action the agent took.
• The result or error.
• Any approval that is still waiting for review.

​

Meetings

Open the agent’s Meetings area to join calls and review recordings, transcripts, and notes. Follow Meetings to configure attendance and recap delivery.

​

Set up each channel

Connect the accounts the agent needs, then set its read and send permissions.

• Email: connect the mailbox.
• Slack: connect the workspace and choose notification channels.
• WhatsApp and Phone and SMS: connect the sending account or number.
• Telegram: configure the agent’s connected Telegram account and message permissions.
• Meetings: connect the calendar and choose joining rules.

​

Use workflows and scheduled work

Use a workflow when the same event or schedule should start a defined process. An AI step can interpret a request or prepare a draft; the following steps determine whether it creates a task, asks for approval, or sends a message. Review those steps and their failure or timeout paths before enabling the workflow. For a message that must wait for a person, follow Approve a sensitive message. Dedicated automatic actions have their own behavior, so do not assume every workflow send asks again at runtime.

​

Common problems

​

Related articles

• AI Hub — configure behavior and knowledge.
• Fix wrong AI information.
• Workflows — recurring processes and run history.
• Artifacts — find files and reports created by the agent.