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

# How Henry works

> The handful of concepts that explain everything Henry does: turns, reads and writes, approvals, memory, tasks, and credits.

## A turn

Every time you ask Henry something, it runs one **turn**: it reads your message, decides which tools to call, calls them, thinks, and answers. A turn can involve many tool calls. Henry shows what it is doing while it works and answers in the same thread you asked in.

Turns have a time budget. Most finish in seconds. A few minutes is the ceiling for an interactive chat turn; when a job needs longer, Henry hands the rest to a background run and posts the result in the same thread. See [Long-running work](/user-guide/working-with-henry/long-running-work).

## Reads and writes

Henry's tools come in two kinds.

* **Reads** look things up: search your inbox, list open tickets, pull a report. Reads run immediately.
* **Writes** change something in the world: send an email, create a ticket, update a spreadsheet, charge a card. By default a write **stages a confirmation** and waits for you.

That default is a setting, not a rule. For any tool you can choose **Run automatically**, **Ask for confirmation**, or turn it off, either from the buttons on a confirmation card or from the Integrations page. The one exception: anything that spends money always asks. See [Approvals](/user-guide/working-with-henry/approvals).

## Channels

A channel is a place you talk to Henry: Slack, Microsoft Teams, Messages on your phone, the web app, and the iPhone and Mac apps. It is one Henry with one memory. Something you tell it in Slack is known when you text it later.

Slack and Teams are connected for the whole workspace. Texting is linked per person. See [Channels](/user-guide/channels/overview).

## Memory

Henry remembers decisions along with the date and the thread they happened in, the people it works with and their preferences, and the dates that matter. Memory is scoped to your workspace and never used to train models. When two memories conflict, Henry trusts the newer one and checks old time-sensitive facts before relying on them.

Some memory is learned passively from channels Henry has been invited to. You can turn that off per channel or purge what Henry learned from a channel at any time. See [What Henry remembers](/user-guide/memory/what-henry-remembers).

## Integrations and connections

An **integration** is a tool Henry can use, like Gmail or Linear. A **connection** is one signed-in account for that tool. A workspace can have several connections to the same tool, each with its own sharing rules:

* **Team-only** connections are shared with everyone in the workspace.
* **Private** connections are usable only by the person who connected them and people they invite.

When Henry says an integration is private to someone else, the fix is to ask that person for access or to connect your own account. See [Integrations](/user-guide/integrations/overview).

## Tasks

A **task** is work Henry does on a schedule: once, every few minutes, hourly, daily, on weekdays, or weekly. Tasks are approved once, at activation, and then run unattended. They report back into the conversation they were created in. See [Tasks](/user-guide/tasks/overview).

## Credits

Henry meters work in **credits** rather than charging per seat. Everyone in a workspace shares one balance, every plan includes unlimited teammates, and small turns are free. Every request has a hard ceiling, so a single task cannot drain your balance. Usage is visible to the whole workspace on the Usage page. See [Billing and credits](/user-guide/admin/billing-and-credits).

## Roles

Every workspace has an **owner**, any number of **admins**, and **members**. Admins manage members, integrations policy, billing, chat connections, and security settings. Members use Henry. The owner cannot be removed. See [Members and roles](/user-guide/admin/members-and-roles).