> For the complete documentation index, see [llms.txt](https://shoppad.gitbook.io/yedric/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://shoppad.gitbook.io/yedric/going-further/conversations.md).
# Conversations
Conversations is the central hub for monitoring and analyzing all user interactions with your assistant. From a straightforward chronological list to in-depth reporting via Trends, Conversations gives you full visibility into how users are engaging with your product.
This view presents all user conversations in chronological order, from newest to oldest. Each row shows a preview of the first user message, the date, the number of turns, a sentiment badge, and any tags on the conversation.
Use the filters at the top of the list to narrow it down. You can filter by username, by tag, by date range, and by turn count.
**Group by user**, at the end of the filter bar, is a view toggle rather than a filter. Turn it on and each user's conversations sit together under a header showing their username and a count, with groups ordered by their most recent activity, so someone asking about the same problem across several threads reads as one block. Conversations with no username collect under **Anonymous visitors**. It is off by default, is carried in the page URL so it survives navigating into a conversation and back, and is left alone by **Clear all**. Click a username in a group header to filter the list to that user.
The date filter follows your own time zone. Picking a day means that day where you are, not in UTC, so the rows you get back match the dates you picked. Note that the day buckets on the Trends view are still calculated in UTC, so a Trends chart and a filtered list can disagree slightly at the edges of a day.
## Search
The keyword search box searches across the full text of your conversations, not just the preview. It matches on message content, the AI-generated summary, the first-message preview, detected topics, tags, and any surfaced feature requests or pain points. Matching is case-insensitive. Search combines with the other filters, so you can, for example, search a keyword within a specific date range or tag.
## Conversation Detail
Click any conversation to open the full detail view. This shows the complete back-and-forth between the user and the assistant, including any tools that were called and what they returned.
The header names the page the visitor was on when the conversation began, labelled **Started on** and linked to the page. If they kept chatting while moving around your site, the later pages are collapsed behind **Also visited N pages**; expand it to see them. Only conversations that come through the website widget record a page, and the start page is fixed once written, so it always says where the visitor came in rather than following them around the site. From here you can:
* **Leave an internal note**: Add a comment visible only to your team, not the end user. Type `@` to mention a teammate in a note. Notes are useful for flagging something to follow up on or adding context for a teammate reviewing the conversation later.
* **Add a custom tag**: Label the conversation with any tag that helps your team organize or filter it. Start typing and the field suggests tags your team already uses, so you can pick one instead of retyping it. Matching ignores case, which means `Refund`, `refund`, and `REFUND` settle on a single tag rather than splitting into three. Tags are capped at 40 characters.
* **Star the conversation**: Click the star on any conversation row to flag it for follow-up. Starring is stored as a tag, so a starred conversation can be filtered for like any other tag, and anyone on your team sees the same stars.
* **Download the transcript**: Export the full conversation as a file.
* **Delete the conversation**: Remove it from your records.
## Tags
Conversations can have two kinds of tags.
**Custom tags** are labels your team adds manually from the detail view. Use these however makes sense for your workflow.
**System tags** are applied automatically by Yedric based on what happened in the conversation. These include:
* **Frustrated**: The conversation had a high frustration score, suggesting the user struggled to get what they needed.
* **Human Support**: The user asked to speak to a person, either through the TalkToHuman tool or by asking directly.
* **Bug**: The assistant detected or the user reported a bug.
* **Question**: The conversation was primarily a question-and-answer exchange.
* **Review Requested**: The assistant prompted the user to leave a review.
* **Good Response**: The user upvoted a response.
* **Bad Response**: The user downvoted a response.
You can filter the list by any tag to find conversations matching a specific pattern.
## Automatic tagging and sentiment
The system tags, sentiment badges, and feature-request detection above are produced by an analyzer that runs over each conversation. You can tune its behavior from the assistant's **Settings** tab, in the **Conversation tagging** card:
* **Auto-tag frustrated conversations**: Turn automatic frustration tagging on or off.
* **Frustration sensitivity**: A slider that sets how readily a conversation is tagged Frustrated. A lower threshold is more sensitive and tags more conversations.
* **Detect feature requests**: Turn detection of feature requests on or off.
* **Analyzer instructions**: Optional guidance for the analyzer that refines how it judges sentiment, frustration, and feature requests. This adds to the built-in analysis rather than replacing it.
## Developer, preview, and simulation conversations
Conversations created by your team, by widget previews, or by [Simulations](/yedric/going-further/simulations.md) are separated from real user conversations so they do not skew your reporting. Simulation conversations are hidden from the list, Trends, and insights by default.
The row menu at the top of the list (the **...** button) lets you toggle whether developer, preview, and simulation conversations are included. When shown, developer conversations carry a Developer chip and simulation conversations carry a Simulation chip so you can tell them apart at a glance.
## Voting
If voting is enabled in your widget settings, users can give a thumbs up or thumbs down on any response. Votes appear in the conversation detail view and automatically apply the **Good Response** or **Bad Response** system tag.
## When to Use Conversations
Use this view when you need to review individual interactions, collaborate with your team on a specific exchange, or follow up on a flagged conversation.
