Skip to content

A daily inbox brief with Claude and Koltrix, under 150 words

Build a short, read-only morning brief from Koltrix with one saved prompt: what feeds it, a sample, how to tune it, a Monday version and what it cannot know.

Koltrix Team5 min read
Black and white analog alarm clock
Photo by insung yoon on Unsplash
On this page(8 sections)
  1. What the brief is made of
  2. The prompt
  3. A sample brief
  4. Tuning it
  5. A Monday version
  6. What the brief cannot do
  7. Turning the brief into action
  8. Key takeaways

A good daily brief answers one question in about a minute: what do I need to know before I open my inbox? A bad one is a wall of summaries you now also have to read.

With Koltrix connected to Claude or ChatGPT, you can build the good version from a single saved prompt. This post gives you the prompt, explains which Koltrix tools feed it, and shows how to tune it so it stays short enough to be useful. For the general idea of a morning briefing, see a morning briefing with AI; this one is specific to a connected assistant.

What the brief is made of

The assistant builds the brief from reads. Which tools it uses tells you what it can and cannot know:

Question in the brief Tool behind it
How much is waiting? get_inbox_summary: unread counts overall and per category
What needs a reply? get_inbox_summary: the conversations Koltrix thinks need a reply
What did my important people send? search_mail filtered by sender or domain
What did I star and forget? list_threads in the starred folder
What is actually in this thread? get_thread

Everything here is read-only. The brief changes nothing in your mailbox, and get_thread does not even mark conversations read, so generating it will not clear your unread numbers.

One limit to be honest about: a connected assistant answers when you ask. The brief is something you run, not something that arrives by itself at 7 a.m. Some assistant apps offer their own scheduling features, but that is a feature of the assistant app, not of Koltrix, so check what your app does and what it asks you to approve.

The prompt

Save this somewhere you can paste it in one keystroke:

Make my morning brief from Koltrix. Keep it under 150 words.

  1. Counts: unread in Primary, Other and Cold pitches. Ignore Cold pitches after the count.
  2. Waiting on me: conversations that need a reply, one line each (who, what they want, any date).
  3. Starred: anything I starred that is older than three days.
  4. People who matter: anything new from acme.com or from Dana.
  5. End with the single thing I should answer first, and why.

Do not draft, label, archive or change anything. If you are unsure whether something needs a reply, put it under "maybe".

Each line is there for a reason.

  • "Under 150 words" is the most important line. Without a limit the assistant explains everything.
  • "Ignore Cold pitches after the count" keeps the strangers who are selling to you out of your attention while still telling you the pile exists.
  • "Do not … change anything" makes the brief a pure read. Rereading it later will not surprise you.
  • "The single thing I should answer first" forces a decision, which is what a brief is for.
  • "Maybe" gives uncertain items somewhere to go.

A sample brief

The shape you want looks like this (the names and mail are illustrative):

Unread: Primary 6, Other 14, Cold pitches 9.

Waiting on you
- Dana (Acme): can we ship Friday? Wants an answer today.
- Priya (accounting): needs the signed renewal by the 14th.
- Sam (vendor): asking about invoice 2041, no date.

Starred, over three days old
- Kickoff notes from Lena.

New from people who matter
- Dana (above); nothing else from acme.com.

Maybe: a customer thanking us; could be a testimonial request.

Answer first: Dana, because a shipping date is waiting on you.

Under 150 words, and the first thing you do after reading it is clear.

Tuning it

Run it for a week and adjust.

If it is too long, lower the word limit and cap each section: "at most five lines per section." Briefs creep.

If it misses things, check the assumption before blaming the prompt. Does Koltrix list it as needing a reply? That list is Koltrix's judgment, and the assistant starts there. If something slipped through, ask a follow-up: "Search the last 48 hours of Primary for any question addressed to me that is not on your list."

If it is wrong about urgency, tell it your rules once: "A message mentioning a date within two days is urgent." Put that in the saved prompt.

If you use several mailboxes, say which. list_mailboxes shows the addresses you can open, and a prompt like "only the support@ mailbox" stops the brief from mixing a shared inbox with your personal one. Shared mailboxes and MCP covers the team angle.

A Monday version

Mondays want a wider look back. Same structure, different window:

Make my weekly brief. Look at the last seven days. Count conversations I replied to versus ones still waiting. List the three oldest conversations waiting on me, and anything I promised a date for that has passed. Under 200 words. Change nothing.

The assistant can use search with date operators for that window, since Koltrix search understands after: and before:. You do not have to write them; describe the window in words.

What the brief cannot do

  • It does not know your calendar unless your assistant also has a calendar connected. A brief about "what is due today" only knows what the email says.
  • It is not complete by promise. The assistant works from Koltrix's own triage and from searches. Treat it as a very good map and skim the inbox if the stakes are high.
  • It can misread. An unusual thread may be summarized wrongly. Open the thread before you act on anything that matters.
  • It reads untrusted text. Mail bodies are written by other people. Koltrix fences them and marks them as third-party content, and the assistant is told to treat that content as data. If a brief ever tells you to do something because an email said so, read the email yourself.

Turning the brief into action

The brief is the read half. When you are ready for the write half, use a second prompt and keep it separate:

From the brief, star the ones I should answer today. Label the "maybe" ones "Check". Then draft a reply to Dana: yes on Friday, ask for the PO number. Do not send anything.

Two prompts, two decisions. The first never changes anything, so you can run it freely. The second acts, and you read what it did. Organizing is reversible, and the draft waits in your Drafts folder with a link that opens it in Koltrix. Sending is off by default and needs your yes if an admin enables it.

Key takeaways

  • A good brief is short: set a word limit, and say what to leave out.
  • The brief is read-only. It uses get_inbox_summary, searches, and get_thread, and it does not mark anything read.
  • A connected assistant runs the brief when you ask. Scheduling, if any, comes from the assistant app, not from Koltrix.
  • Save the prompt, tune it for a week, and keep the "act on it" prompt separate from the "read it" prompt.
  • More prompts like this are in triage with Claude, and the overview is at koltrix.com/mcp.

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.

SharePost on XLinkedIn
More in Guides →