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

# Test a procedure before setting it live

> You can try a procedure three ways before customers meet it: chat with it, check which recent conversations its When to use picks, and replay it on a past conversation.

You can try a procedure three ways before customers meet it: chat with it, check which recent conversations its **When to use** picks, and replay it on a past conversation. Open the procedure under **Procedures** and choose **Test**. The panel has three tabs: **Preview**, **Trigger** and **Replay**.

Nothing you do in **Test** reaches a customer. A procedure that has never been set live runs only there.

To test the whole agent instead of one procedure, see [Test your agent before going live](https://docs.aide.app/aide-ai-agent/test-agent.md).

## Preview: chat with the procedure

**Preview** opens a chat with your agent that starts in this procedure. Write as a customer would and follow the steps to the end.

* The chat follows the procedure from the first message, even while the procedure is paused or its **When to use** wouldn't match. It tests the steps, not the trigger.
* It runs what is saved, with your draft if you have one. With unsaved edits, the panel asks you to **Save** first.
* **Reset** starts a new chat.

## Trigger: check which conversations it picks

**Check trigger** scores the **When to use**, with your unsaved edits, against the last 50 conversations in the places the procedure runs.

* The result reads, for example, "Matches 6 of the latest 50 conversations", with the threshold it was scored at.
* **Matched** and **Declined** list the conversations, each with its score. Select one to open it.
* A near miss, or a match Aide had doubts about, comes with a short reason.
* After you edit the text, the result says the trigger changed. Choose **Check again**.

The same check is under **When to use** in the editor, next to **Train on examples**. There, each conversation with a reason has a button:

* **Use for this** on a declined conversation adds its message to **Use when**.
* **Don't use for this** on a matched conversation adds it to **Don't use when**.

Each list holds 20 examples (see [Choose when a procedure runs](https://docs.aide.app/automation/procedure-when-to-use.md)).

## Replay: run it on a past conversation

1. Open **Replay** and search your recent conversations.
2. Select one. Aide drafts a reply to that conversation's latest customer message, with the saved procedure.
3. Read the result: whether Aide would reply, hand the conversation to your team or not reply, the reply it wrote, and whether this procedure applied. If it didn't fit the message, the result says so and names the procedures that did apply.

The result ends with "Nothing was sent." Replay runs one conversation at a time, for this procedure. It doesn't replay your past conversations against the whole agent.

You can also replay from the editor: the example conversations listed under the **Only if** conditions each have a preview button.

## Limits and cost

* **Check trigger** and **Replay** use your workspace's AI budget. When today's budget is spent, they say so and you can try again tomorrow.
* A workspace can run 20 trigger checks and 30 replays an hour.

## Who can do this

**Test**, **Check trigger** and replays are for owners and admins.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.