AI Agent

Documentation Index

Fetch the complete documentation index at: https://docs.trellistech.com/llms.txt

Use this file to discover all available pages before exploring further.

​

Meet your AI teammate

The Trellis AI agent is like having an experienced team member available inside your connected conversations and Trellis workspace. When an agent is assigned and AI is enabled for the conversation, it uses the property, reservation, time of day, instructions, knowledge base, and conversation history to draft or send accurate, helpful answers. But it also knows its limits. When something is outside its scope - a complaint, a refund request, a safety issue - it escalates to your team instead of guessing.

---

​

What the AI agent can do

---

​

How the AI responds to guests

---

​

Reservation lookups

When the AI agent looks up reservations - for reports, guest questions, or operational queries like “tomorrow’s checkouts” - it automatically filters to active reservations only. This means results include confirmed, checked-in, and checked-out stays, while inquiries, cancelled bookings, and no-shows are excluded by default. This prevents unconfirmed or cancelled reservations from appearing in operational reports and guest-facing answers, so your team only sees bookings that matter.

---

​

Property lookups

The AI agent is aware of where each property sits in your pipeline - Lead, Evaluation, Onboarding, Active, Paused, or Churned. When it searches across your portfolio for day-to-day operational questions like “how many properties do we have?” or “what’s near the airport?”, it focuses on Active properties only by default. Leads, properties still in onboarding, paused listings, and churned properties are left out unless you ask for them specifically. This keeps operational reports, guest-facing answers, and routine queries focused on the listings that are actually live and taking bookings.

---

​

Availability search

When a guest asks about open dates - “do you have a place near the beach for next weekend?” - the AI agent searches active properties for availability on the requested dates. It returns matching options with property details such as location, bedrooms, bathrooms, and guest capacity. For large portfolios, the agent evaluates the candidate properties before applying the reply limit. If more properties match than can be shown in a single reply, the agent tells the guest how many total results were found so they can refine their search.

---

​

Custom fields

The AI agent can use reservation-level custom fields that have synced into Trellis. This means the agent can reference available custom data - like access codes, special requests, or booking-specific notes - when answering guest questions. You do not need to configure anything extra. As long as your PMS integration syncs custom fields into Trellis, the AI agent automatically includes them in its context when responding to guests.

---

​

Slack file reading

When your team shares files in a connected Slack channel - documents, spreadsheets, or images - the AI agent can read and understand them. Instead of just seeing a file name, the agent downloads the file, reads its contents, and uses that information to answer questions or take action.

• Text files - the agent reads the full content of text-based files (documents, CSVs, logs) and can reference specific details in its responses.
• Images - the agent describes what it sees in uploaded images, which is useful for maintenance photos, guest ID documents, or property screenshots.

---

​

Setting up your AI agent

​

Writing great agent instructions

Your agent instructions tell the AI how to represent your business. The better your instructions, the better the AI performs. Include in your instructions:

​

Configuring autopilot

You control how much autonomy the AI has for each type of contact:

---

​

Connected accounts

Connected accounts let your AI agent log in to external websites on your behalf - OTA portals like Airbnb and Booking.com, PMS dashboards, cleaning platforms, and more. Once connected, the agent can browse these sites to look up information, check claim statuses, download reports, or take actions that are only available through a web browser.

​

How it works

You save your login credentials for each website, and Trellis securely stores and encrypts them. When the AI agent needs to access that site, it opens a cloud browser, logs in using your credentials, and completes the task - just like a team member would.

​

Adding a connected account

​

Connection statuses

​

Re-verifying an account

If an account status changes to Failed or Needs code, click Re-verify on the account card. Trellis runs the automated login again. If the account was previously working (status was Ready), Trellis starts the re-verification automatically. If it fails, you will be prompted to open the live browser view and resolve the issue manually.

---

​

Knowledge base

The knowledge base gives your AI agent access to documents and information beyond what is in your agent instructions. What to add to the knowledge base:

• Property-specific guides (detailed check-in procedures per property)
• Local area guides and recommendations
• Seasonal information (pool opening dates, ski pass details)
• Cleaning and maintenance procedures your team follows
• Standard operating procedures for common situations

---

​

Agent memories

Over time, your AI agent learns from interactions and builds a memory of important details. Memories help the AI give more consistent, personalized responses. You can review, edit, and delete memories at any time from the Knowledge Base section.

---

​

Multiple agents

For larger operations, you can create multiple AI agents - each with different instructions, permissions, and areas of focus. Common setups:

---

​

Google Meet bot

Your AI agent can join Google Meet calls and respond to questions live - using the same knowledge base, workspace access, and instructions as your text-based agent.

​

How it works

The bot joins a Google Meet call as a participant and listens for its wake word. When someone says the wake word, the bot processes what was said and responds with natural speech.

• Wake words - the bot starts in a dormant state and activates when someone says your agent’s first name. By default this includes the name on its own and “hey” followed by the name (for example, “Lucy” or “hey Lucy” if your agent is named “Lucy - Property Manager”). “Trellis” is always available as a universal fallback, even after you customize the list.
• Dismiss words - when you are done talking to the bot, say “bye” or “goodbye” followed by the agent’s name (for example, “bye Lucy” or “goodbye Lucy”). The bot returns to its dormant state and stops responding until the next wake word.
• Custom wake and dismiss words - you can add your own activation and dismissal phrases on the meetings settings page. The defaults stay in place unless you remove them.
• Full workspace context - the bot can look up reservations, check property details, search tasks, browse your knowledge base, and more - just like the text-based agent.
• Meeting-aware - the bot understands that multiple people may be speaking and responds to whoever addressed it.
• Conversation memory - everything discussed during the call is saved to the agent’s session history and to a Meetings folder in your knowledge base, so you and the AI can reference it later.

​

Setting up the Google Meet bot

​

Bot identity in calls

When the bot joins a meeting, its video tile displays the agent’s name and current status so participants always know who it is and what it is doing:

• Connecting - the bot is joining the call
• Dormant - the bot is waiting for its wake word (shown as a muted gray indicator)
• Listening - the bot is active and listening to the conversation
• Speaking - the bot is responding to a question
• Error - something went wrong (check your agent configuration)

The Meetings page shows how many times the bot has been activated, so you can track usage across your team.

---

​

The AI chat sidebar

Beyond handling guest messages, you can chat directly with the AI agent from anywhere in Trellis. Use the AI sidebar to:

• Ask questions about your properties, reservations, or guests
• Get quick summaries of conversations or tasks
• Create tasks, add comments to task activity feeds, or update records by simply asking
• Manage departments - create new ones, rename existing ones, or remove ones you no longer need
• Build or edit task templates by describing what you need
• Review workflow approval context and discuss next steps before approving or rejecting
• Draft messages for manual review
• Pull finance reports - PM commission, revenue by listing, taxes, and fees
• Get operational insights and recommendations

​

Switching AI modes mid-conversation

You can switch between AI quality modes at any point during a conversation - even after messages have already been sent. Open the mode selector in the chat input area and choose a different mode. A confirmation appears so you know the switch was applied. Each response shows which mode was used, so you can compare how different modes handle the same type of question.

​

Activity details

While the AI is working on your request, the chat sidebar shows its activity in real time. You can see the operational steps it is taking - such as looking up a reservation, searching your knowledge base, or checking property availability - as each action happens. Actions appear in chronological order, so you always know what the AI did and in what sequence. Click any action to expand the supporting details.

​

Entity references

When the AI mentions a contact, conversation, memory, knowledge base document, or skill in its response, the reference appears as a clickable pill. Click any pill to open a detail panel on the side - you can see the full record without leaving the chat. Skills referenced in the thought-process timeline are clickable too, so you can jump straight from a reasoning step into the underlying knowledge base document. Each entity type shows relevant details:

​

Inline entity lists

When the AI returns a list of records - for example, “show me today’s check-ins” or “which tasks are overdue?” - the results render directly in the chat as a tidy, clickable list instead of plain text. Each row uses the same layout you see on the corresponding page in Trellis, so a property in chat looks like a property on the Properties page, and a task in chat looks like a task in the Tasks list. Inline lists are supported for these record types: Click any row to open the full record in the side panel - the same drill-in experience you get from entity pills, including the breadcrumb trail when you go more than one level deep.

​

Drilling into related records

Detail panels are connected. When you open an entity and click through to a related record - for example, opening an action result, then drilling into the user it references, then jumping to one of their tasks - each record stacks inside the same panel instead of opening in a separate sidebar alongside the first one. A breadcrumb trail appears at the top of the panel as soon as you go more than one level deep. It shows the path you took so you can retrace your steps at a glance:

• Click any crumb in the trail to jump straight back to that record.
• Click the back arrow to return to the previous record.
• Press Escape to step back one level, or close the panel if you are at the top.

If your trail grows longer than a few levels, the middle crumbs collapse into a … menu so the header stays readable. Click the menu to see every step you took and pick one.

​

Managing chat tabs

Chat tabs show both an Archive button and a close button when the conversation has messages. Use Archive to save the conversation for later review, or close to dismiss the tab without archiving.

---

​

Pulse - AI activity feed

Pulse gives you a real-time view of everything your AI agent is doing across your workspace. Instead of checking individual conversations or tasks to see what the AI handled, Pulse shows a unified feed of every action - tasks created, messages sent, workflows triggered, and more.

​

Where to find Pulse

Pulse appears in two places:

• Dashboard - the Recent AI activity section shows a feed of the most recent actions your AI agent has taken across all conversations and properties.
• Conversation sidebar - the Pulse tab on any conversation shows AI activity scoped to that specific conversation, including actions on related tasks and reservations.

​

What Pulse shows

Each row in the feed represents an action your AI agent performed. Common actions include: When the AI performs the same type of action multiple times in a single run - for example, creating several tasks at once - those actions are grouped into a single row that reads “Created 4 tasks” instead of showing four separate entries. This keeps the feed easy to scan.

​

Viewing activity details

Click any row in the Pulse feed to open a detail panel showing:

• Started by - what started the AI run (a guest message, a Slack message, a scheduled workflow, or a voice session). Click the source to jump to the original conversation or workflow.
• Activity details - the steps the AI took and the workspace data it used. This is the same activity view you see in the AI chat sidebar.
• Response - the final message the AI composed, rendered with clickable entity references so you can verify the sources the AI cited.
• Affected objects - preview cards for each task, property, reservation, or conversation that the action touched. Click any card to open its full detail view.

​

Failed actions

If an AI action fails - for example, the agent tried to create a task but could not complete the operation - the row appears with a red indicator and an error summary. For grouped actions with partial failures, the feed shows how many succeeded and how many failed (for example, “3 of 5 failed”).

​

Sharing an activity

Each detail panel has a unique URL. Copy the URL from your browser’s address bar to share a specific AI activity with a team member - they will see the same trigger, details, and response when they open the link.

---

​

Voice meetings

Your AI agent can join video meetings and participate using real-time voice - just like having a team member on the call. When you invite the AI agent to a meeting, it joins the video call and listens to the conversation in real time. It can respond naturally with speech, answer questions, and provide information - all using the same property knowledge, reservation context, and agent instructions it uses for text conversations. After each meeting, a summary of what was discussed is automatically saved to the Meetings folder in your knowledge base. This means the AI agent can reference past meeting context in future conversations - so if something was decided on a call, the agent already knows about it.

​

The Meetings workspace

Every call the AI attends shows up in a dedicated Meetings workspace in your knowledge base. Open Knowledge Base > Agents, select your agent, and click the Meetings tab to see a full history of past and upcoming meetings. The list is organized by status tabs so you can quickly focus on what matters:

​

Meeting detail page

Click any meeting to open its detail page - a complete record of the call that includes notes, the full transcript, participants, and anything the AI linked while listening. The layout adapts based on the meeting’s state so you always see the right information at the right time:

• Upcoming - a “Starts in…” hero with the invited participants and a preview of what the agent plans to cover
• Live - a video hero with the active speaker pill, an in-meeting participants panel, and a live activity feed of what the agent is doing
• Processing - a step list that shows what the agent is still working on (transcript, notes, summary), plus an early summary as soon as it is ready
• Done - the full recap layout with video, notes, transcript, and sidebar insights
• Failed - a simple centered hero explaining why the call could not be completed

On a completed meeting, you will see the following sections:

​

Meeting recap emails

After each completed meeting, Trellis can automatically email a recap to the attendees. The recap includes the meeting title, date, a clean summary of what was discussed, and a link to the full meeting detail page. Recap emails are turned on per agent in the meetings settings page. You choose which meetings trigger a recap - every meeting, only meetings the AI actively participated in, or only meetings that produced notes.

​

Calendar-scheduled meetings

Connect your agent’s calendar from Meeting Settings to have it join meetings automatically. Once connected, any event on the calendar with a supported video link (Google Meet, Zoom, or Microsoft Teams) shows up in your Meetings workspace as a scheduled meeting. Trellis dispatches the agent to the call at the scheduled time - you do not need to invite it manually.

​

Joining preferences

Choose which kinds of events your agent should join using four simple toggles. Turn on any combination - the agent joins an event when any enabled toggle matches. By default, new connections have External, Not organized by me, and Not accepted turned on, and Team turned off - so the agent covers customer-facing and inherited meetings without sitting in on every internal standup. Adjust the toggles at any time to match how your team uses calendars.

​

Per-event attendance overrides

Sometimes you want the agent to attend a meeting it would normally skip, or skip one it would normally join. From any scheduled meeting in the Meetings list - or directly on a calendar event - you can override the default joining preferences for that single event:

• Force attend - send the agent to a meeting that does not match your joining preferences (for example, an internal standup you would normally skip)
• Skip this one - keep the agent out of a specific meeting without changing your overall preferences
• Reset to default - clear the override and let your joining preferences decide again

Overrides apply to a single event and stay in place even if the event is updated on your calendar. Recurring events can be overridden one occurrence at a time, so changing your mind about today’s standup does not affect tomorrow’s.

​

Calendar view

Switch the Meetings list to Calendar view to see scheduled meetings on a month, week, or day grid alongside your existing calendar events. Each event shows whether the agent is set to attend, skip, or follow your default preferences - and you can change the override directly from the calendar without leaving the page.

​

Meetings settings

Each agent has its own meetings settings page, reachable from Knowledge Base > Agents > [Agent] > Settings > Meetings. You can also jump straight there from the gear icon in the top-right of the Meetings list toolbar - a one-click shortcut from the meetings you are looking at into the settings that govern them. From the settings page you can:

• Connect or disconnect a calendar
• Set joining preferences for scheduled meetings
• Review and clear per-event attendance overrides
• Customize the wake and dismiss words the bot listens for during calls
• Toggle recap emails on or off and choose who they go to
• Enable or disable automatic note enhancement after each call
• Configure which workspaces and teams can see meeting notes

​

Meeting reports

After each meeting, the AI agent automatically saves a report to your knowledge base in a Meetings folder. The report captures what was discussed during the call so your team can reference it later. Meeting reports are shared with all agents in your workspace, so any agent - whether it handles guest messages, owner updates, or internal operations - can access past meeting context when answering questions.

---

​

Connect your agent to more channels

Beyond the unified inbox, your AI agent can communicate through additional connected channels. Each enabled channel works the same way - the agent reads messages, gathers context, and responds or escalates based on your instructions and channel settings.

​

Email

Give your agent a personal email address. Anyone who emails it - guests, vendors, partners - gets an AI-powered response using your agent’s knowledge and instructions. You can also forward emails to the agent for processing.

​

Slack

Connect your agent to a Slack workspace so your team can message it directly in channels or DMs. The agent responds in-thread, keeping conversations organized.

​

WhatsApp

Give your agent a WhatsApp number so guests can message it directly. The agent handles the conversation the same way it handles inbox messages - using your instructions, knowledge base, and escalation rules.

​

SMS

Connect a phone number for SMS conversations. Guests who text the number get AI-powered responses.

​

Telegram

Connect a Telegram bot so your agent can respond to messages on Telegram. Each channel integration is configured in the Integrations tab of your agent settings. The agent uses the same instructions, knowledge base, and memories across supported connected channels - so a guest gets the same quality response whether they message on WhatsApp, email, or Airbnb.

---

​

Workflows

Your AI agent can build and run automated workflows - multi-step sequences that handle repetitive operational tasks without manual intervention. There are two ways to create workflows with the AI:

• From the Automations page - go to Automations, click New Automation, describe what you want in the “Describe what this workflow should do…” input, and click Build. The AI chat sidebar opens seeded with your description, asks any clarifying questions it needs, and assembles the complete workflow for you.
• From the chat sidebar - describe what you need in plain language directly from the AI chat sidebar, anywhere in the app. For example: “Every morning at 9am, check today’s checkouts and create a cleaning task for each one.” The agent builds the workflow with the right triggers, steps, and conditions.

Workflows can include:

• Queries - pull data from reservations, tasks, properties, or contacts
• Conditions - branch logic based on data (if checkout is today, if task is overdue)
• Actions - create tasks, update records, send notifications
• Approvals - pause and wait for a team member to approve before continuing
• Schedules - run daily, weekly, or on specific triggers

Manage your workflows from the Automations page in the sidebar. Each workflow shows its run history, current status, and any pending approvals.

---

​

Scheduled tasks

Your AI agent can schedule tasks for itself - things it should do later, on a recurring basis, or when a specific event happens. Examples:

• “Remind me to check Haven Beach House reviews every Friday”
• “Send me a daily summary of yesterday’s check-ins at 9am”
• “When a new reservation is created, look up the guest and add a welcome note”

The agent tracks its scheduled tasks and runs them at the right time. You can view and manage all scheduled tasks from the agent’s settings.

​

What you see when a scheduled task runs

Every time a scheduled task fires, Trellis surfaces the run in the three places you already check on the agent - so you never have to go hunting for what happened.

---

​

Tips for getting the best results