> ## Documentation Index
> Fetch the complete documentation index at: https://docs.gravitygtm.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Using Horizon in Slack

> Compound requests, list processing, spend confirmation, and daily limits

This page covers what Horizon can do once it's [installed and signed
up](/horizon/slack-install) in your workspace, and the guardrails that keep it
predictable to use.

## Talking to it

* **@mention** it in a channel: `@Horizon ...`
* **DM** it directly — no @mention needed
* **Slash command**: `` `/usehorizon <your question>` ``
* **Follow-ups in a thread** it's already replied in don't need a fresh
  @mention — Horizon reconstructs the thread's history, so "now find their
  emails" works on the very next message and knows exactly who "their"
  refers to.

## Compound requests

Ask for something that takes several steps and Horizon chains the
underlying tool calls itself in one turn — you don't have to break it into
separate questions:

> "Find companies like Acme Corp, then find the CEO's email at each one."

Behind the scenes this runs an enrich → derive signals → search → enrich →
find-email chain automatically, with a short progress update in-thread if
it takes a few steps.

## Lists and batches

Paste (or type) a list of names, companies, or URLs and Horizon processes
each independently, then compiles **one combined answer** — never a reply
per item:

> "Enrich this list and find each person's email: \[30 names]"

You can also **attach a file** instead of pasting — a `.csv`, `.tsv`, or
plain `.txt` list of names/companies/URLs is read directly, so Horizon
knows the exact item count up front rather than estimating from a long
paste (this feeds a more accurate credit estimate into the confirmation
step below).

If the result is a tabular list beyond about 20 rows, Horizon delivers it
as a downloadable CSV file attached to the reply instead of a wall of text
in the message itself.

## Honest answers, live cost transparency

Horizon never fabricates a name, title, or email — an honest "not found"
is preferred over a plausible-looking guess, and every answer that spent
credits ends with exactly what it cost:

```
_Spent 27 credits · 173 remaining_
```

## Spend confirmation

Before running anything estimated above **50 credits or 10 items**,
Horizon pauses and posts a card instead of just running it:

> **Confirm before I proceed**
> *(plan summary + credit estimate)*
> \[Run it] \[Cancel]

Clicking **Run it** starts the approved plan; **Cancel** spends nothing.
This is a hard rule, not a suggestion the model can talk itself out of — a
big batch or an expensive multi-step chain always stops here first.

## If a batch is too large for one turn

Very large batches (upwards of \~100 items) can hit a per-conversation
budget before finishing. Horizon delivers whatever it completed so far and
offers to continue — just reply **continue** in the same thread and it
picks up where it left off using the thread's own history.

## Free-tier daily limit

A free-trial workspace gets **20 agent messages per day** (resets at
midnight UTC) — this protects against runaway usage while you're still on
the \$0 trial. Hitting it shows a friendly limit message, not an error:

> You've hit today's free-tier limit of 20 questions — it resets at
> midnight UTC. Upgrade to a paid credit pack any time for unlimited daily
> messages.

**A workspace on any paid credit pack is never capped** — only the request
rate and credit balance shared with the REST API and MCP server apply. See
[Credits & Pricing](/horizon/credits-and-pricing).

## Out of credits

If a request runs out of credits mid-conversation, Horizon says so plainly
and includes a top-up link — it never fails silently or hands back a
partial result without explaining why.