Five email agents to set up in minutes, all starting on Suggest
Five email jobs to hand an agent first: triage, invoice chasing, lead qualifying, scheduling and follow-ups. Each starts on Suggest, so you review every draft.
On this page(10 sections)
- What does "starts on Suggest" mean?
- The five recipes at a glance
- 1. How does a support triage agent help?
- 2. How does an invoice collector agent work?
- 3. What does a lead qualifier do?
- 4. What can a meeting scheduler agent do?
- 5. How should a follow-up chaser behave?
- In what order should you turn them on?
- A ten-minute checklist before installing any recipe
- Bottom line
The five email jobs most worth handing to an agent first are support triage, invoice chasing, lead qualification, meeting scheduling and follow-ups. They are repetitive, the replies are short, and a wrong draft is cheap to catch. Koltrix ships each as an agent recipe that you install in one click, and every one starts on Suggest.
An email agent is a program, usually driven by an AI model, that reads mail sent to its own address and acts on it. In Koltrix that address is an agent mailbox: a real address on your domain, like [email protected], with a key that reaches that one mailbox and nothing else. This post covers what each of the five does, what to watch in its first drafts, and the order to turn them on.
What does "starts on Suggest" mean?
Suggest is the default setting on the autonomy dial: the agent writes drafts and proposals, and a person reads them and sends. The agent has no way to send on its own. The next level up, Approve, lets the agent propose a send that waits in your Action center until you approve, edit, reject or snooze it. The top level, Auto (Beta), lets an agent reply on its own when it is sure. It is off by default for every agent, sends only when highly confident and grounded, to allowlisted recipients it already knows, inside a daily cap, and logs every send.
Installing a recipe also sets the injection shield to quarantine. Every inbound message is scanned for prompt injection, and a high-risk one is held back from the agent and shown to a person with the reasons.
The five recipes at a glance
| Recipe | What it is for | What to review in the first drafts |
|---|---|---|
| Support triager | Answers the first email so you do not have to | Does it admit what it does not know instead of guessing? |
| Invoice collector | Chases unpaid invoices, politely and on time | Amount, invoice number and tone |
| Lead qualifier | Reads new enquiries and asks the right questions | Are the questions short and relevant to your product? |
| Meeting scheduler | Finds a time that works for everyone | Time zones and the times it offers |
| Follow-up chaser | Nudges people who have gone quiet | Does it stop when the person has replied? |
Each recipe comes with its own mailbox, instructions and rules, and each counts as one agent mailbox against your plan. The instructions are yours to edit, so treat the recipe as a starting point and not a finished policy.
1. How does a support triage agent help?
A support triage agent drafts the first reply to a new support email, so a person starts from a draft instead of a blank box. It earns trust fastest because the cost of a bad draft is one edit.
Watch for three things in the first week:
- Confident guesses. A good first reply says what the agent knows and what it will check. Add a line to the instructions telling it to say "I'll find out" when the answer is not in what you gave it.
- Tone drift. Compare the drafts with a few of your own sent replies. If they sound different, put two or three of your real replies into the instructions as examples.
- Wrong routing. If it keeps drafting answers to billing questions that only you can resolve, tell it to hand those to a person instead of drafting.
If you run a shared inbox, see triage rules for shared inboxes for the categories worth separating first.
2. How does an invoice collector agent work?
An invoice collector writes the reminder emails about unpaid invoices that nobody enjoys sending. The value is consistency: reminders go out when they should, in a calm voice, without waiting for you to remember.
Money emails need a stricter review than most, so keep it on Suggest for longer than the others. Check each draft for the right amount, the right invoice reference and a clear next step. Keep the tone factual: a reminder states what is owed and since when, and offers a way to pay or to say there is a problem.
The cadence is your call. Public guides disagree on the exact days, so decide yours, write it into the instructions and adjust for clients who always pay on time. Invoice and receipt emails covers what each message should contain.
3. What does a lead qualifier do?
A lead qualifier reads a new enquiry and replies with the questions you would have asked anyway: what the person is trying to do, how big the team is, and when they need it. The point is to turn a vague "tell me more" into something you can act on.
Keep it to two or three questions. A long questionnaire reads as a form, and people stop replying to it. Allowlisting matters here, because leads are strangers by definition. Public enquiries are the most exposed address you will give an agent, which is why the quarantine default on the shield is worth keeping. A lead that asks your agent to ignore its instructions should land in front of a person, not in front of the agent.
4. What can a meeting scheduler agent do?
A meeting scheduler handles the back-and-forth of finding a time that works for everyone, one reply at a time. It is the most tedious email job and the easiest to check, because a person can see the proposed times at a glance.
The failure to look for is time zones. In every early draft, confirm that the times and the zone names match what you would have written. Meeting requests in a shared inbox lists the details that cause most scheduling mix-ups.
5. How should a follow-up chaser behave?
A follow-up chaser nudges people who have gone quiet after you asked a question or sent a quote. It handles the case where the conversation dies because everyone assumed the other person would reply.
Two rules make it safe to leave running. First, it must stop the moment the person replies, so test that on a real thread. Second, cap the number of nudges in the instructions, for example two, and make the last one a clean close ("I'll assume this isn't a priority right now"). Following up on support without nagging has wording that works.
In what order should you turn them on?
Start with the agent whose mistakes are cheapest and whose output you can judge in seconds.
- Support triager and meeting scheduler first. Drafts are short and easy to check.
- Follow-up chaser second. Test the stop-on-reply behavior before you trust it.
- Lead qualifier third. It talks to strangers, so check the shield's held messages for a week.
- Invoice collector last. It deals with money, so read every draft until you are sure of it.
Run one agent for a week before adding the next. The Activity feed shows what each agent drafted, proposed or held, which is the quickest way to see whether its drafts match your own judgment.
A ten-minute checklist before installing any recipe
- Pick a domain you have verified, and a name that says what the agent does.
- Add a sender allowlist if the agent should only hear from known people.
- Leave autonomy on Suggest and the shield on quarantine.
- Edit the instructions: your tone, your policies, what to hand to a human.
- Send it three test emails, including one that tries to give it an order.
- Know where the kill switch is: one switch puts every agent in the workspace back to Suggest.
For the wider reasoning on which tasks suit an agent at all, read which email tasks are safe to delegate, and for the access side, least privilege for email agents.
Bottom line
Start small: install one recipe, keep it on Suggest, and read what it drafts for a week before adding another. The recipes are a starting point built around common email jobs, not a promise that the agent knows your business. Your instructions, your allowlist and your review are what make the drafts good. If you later try Auto, you will already know which agents have earned it.
Start with Koltrix
Your domain, one inbox, and an API that sends.
A team inbox where AI sorts and drafts (nothing is sent without your click), plus the transactional API and SMTP relay your product sends with. 7 days free, no card.
