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

# Prompt Master

> Author the instructions and reasoning behind an agent.

The **Prompt Master** is where you write what an agent thinks with — its
instructions, structured into clear sections, with variables and a mode that decides
how it behaves at run time.

## What a prompt holds

* **System and user sections** — the standing instructions and the per-run input,
  kept separate.
* **Variables** — typed placeholders you fill in at test/run time.
* **Mode** — `single-shot` (one response) or `chat` (a back-and-forth conversation).
  Mode is a property of the prompt itself.

## Test as you write

The Test panel runs your prompt against any [model](/masters/model) — a
[Viorant tier](/providers/credits) on credits, or one of your own keys — and streams
the response back live. Switch modes to see single-shot vs. chat behavior.

## Comparing versions

Prompts are versioned. When a prompt is in a good state, **commit** a snapshot —
the **Version History** panel then lists every commit, newest first, with its
message and time. From that list you can:

* **See what changed** — open a past commit to bring up its **diff view**. The
  diff isn't a file you edit; it's a read-only view comparing that snapshot to
  the current prompt, with additions and removals highlighted line by line.
* **Restore a commit** — select a version and click **Restore Version**. The Hub
  sets the prompt back to that snapshot and records the restore as a new commit,
  so nothing in the history is lost.

The **current-version** badge marks the latest commit — you can't diff or restore
it against itself. If you have uncommitted edits, they show as a pending change
you can diff against the last commit before deciding to commit or restore.

<Note>
  Screenshots and a full field-by-field walkthrough are being added to this page.
</Note>