Your First Agent

The fastest way to understand TrailKit is to build something useful with it. So let's build a Monday morning inbox digest. It takes about fifteen minutes, and once you've done it the rest of the product clicks into place.

What we're building

Every Monday at 8 AM, the agent will:

1. Read everything that landed in your inbox over the weekend.
2. Pull out what actually needs a reply from you.
3. Lump the rest (newsletters, FYIs, automated stuff) into a short summary.
4. Send you the to-do list and the summary in Slack. Or email, whatever you prefer.

Once you've done one of these, every other agent follows the same shape.

1. Hook up the tools the agent needs

Open Integrations in the dashboard sidebar (the puzzle-piece icon). You'll see a long list of services. For this one we need two:

• Gmail or Outlook, so the agent can read your inbox.
• Slack (or just email) so it has somewhere to send the digest.

Click each, sign in, grant access. You'll bounce through the vendor's normal login. We never see your password. You can revoke the connection from the same panel anytime.

Got two Gmail accounts? Connect both. The agent will ask which one you mean if it isn't obvious from context.

2. Create the agent

Back on the dashboard, click New agent. Name it something future-you will recognize at a glance ("Monday inbox digest" works).

Then write the goal in the chat box like you're briefing an assistant. Don't try to nail the wording on the first pass, you can iterate:

Every Monday morning at 8 AM Pacific, read everything that landed in my Gmail since Friday at 5 PM. Pull out the threads where someone's actually waiting on me, one-line each with a link. Lump newsletters, calendar notifications, and receipts into a short FYI section. DM me the whole thing in Slack.

That's the setup. No fields to fill, no flowchart.

3. Let the agent play it back

The agent will read what you wrote and reply with something like:

• "Got it. Every Monday at 8 AM Pacific I'll search Gmail for messages since the previous Friday at 5 PM. Here's how I'll decide what counts as needing a reply..."
• The connections it'll use (your Gmail and your Slack).
• The schedule it wants to set up.

Read this carefully. It's the agent telling you what it heard. If anything's off, just say so ("use my work Gmail, not personal"). It'll adjust and play it back again.

When it looks right, approve. The agent is live.

4. Watch a run

You can wait until Monday or you can ask for a dry run right now:

Run once now against last weekend so I can see what the output looks like.

The agent runs, posts the digest, and shows you every step: which messages it read, why it classified each one the way it did, what made the cut and what didn't. Every agent has a Runs view where you can scroll back through every wake-up and dig into the details.

If the digest is too long, the wrong shape, missing something important — tell the agent. It'll update its own instructions on the spot. You're not editing a rule file. You're giving feedback to a coworker who remembers.

5. Iterate

The first version is almost never the last one. Stuff people usually add on the second or third pass:

• "Skip anything from no-reply addresses."
• "Treat anything from my CEO as needing a reply, full stop."
• "Add a section for things my chief of staff should see instead of me."
• "If there's nothing urgent, just say 'inbox is clean' and call it a day. Don't pad it."

Each one is a single sentence. The agent absorbs it permanently. You won't have to say it again next week.

What to build next

The second agent is always easier than the first. Some good candidates:

• A calendar tidy-up that flags conflicts, agenda-less meetings, and back-to-backs that need a buffer. See Admin.
• A lead notifier that fires when a form is submitted, looks the person up, and pings you if they're hot. See Growth.
• A weekly metrics summary so you don't have to log into five tools every Monday. See Insights.

Same five steps, different goal.