> ## Documentation Index
> Fetch the complete documentation index at: https://wiredesk.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.
# Conversations
> Every chat, call and email across your agents: how each ended, what it was about, and the full transcript. Take over a chat or email from here.
**Conversations** lists every chat, phone call, browser voice call and email your agents have handled. Each row shows how the conversation ended and what it was about. Open one for the transcript, a summary, and, for chat and email, a reply box so you can take over from the agent.
Everyone in the workspace can use Conversations, whatever their [role](/docs/account/team-and-roles).
## The list
The list shows your 100 most recent conversations, newest first.
| Column | What it shows |
| --------------------- | --------------------------------------------------------------------- |
| **Contact** | Who it was with, the agent that answered, and, for calls, the length. |
| **Channel** | The channel badge, plus a handover badge if a person was asked for. |
| **Outcome** | How it ended, and the customer's sentiment. |
| **What it was about** | The topic, followed by the one- or two-sentence summary. |
| **When** | When it started. |
In the list, each contact is labelled by how they reached you: **Web visitor** for chat, the phone number for calls, the email address for email, **Caller ID withheld** for a call with no number, and **Browser test** for your own voice tests. A single conversation's page shows the person's name, once the agent has learned it.
### Filters
* **Channel tabs:** **All**, **Chat**, **Voice** or **Email**. **Voice** covers both phone calls and browser voice. The badge on each row tells you which.
* **Outcome menu:**
| Option | Shows |
| ----------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Any outcome** | Everything. |
| **Needs you** | Conversations that still need a person: anything still live, any where someone asked for a person (including requests nobody answered), and any that ended **Handed to you** or **Message taken**. |
| **Still open** | Conversations that have not ended yet. |
| **Resolved**, **Handed to you**, **Message taken**, **Abandoned** | Conversations with that outcome. |
* **Search** (**Search by name, topic, or summary…**) matches the name the agent learned for the contact (from a lead, a message it took, or a verified visitor), the contact itself, the phone number as stored (so a pasted `+12018991082` works), the agent's name, the channel, the summary, the topic and the outcome. Where a name is known, the **Contact** column shows it, with the number or address underneath.
Filters apply to the 100 conversations in the list. Links from the dashboard and from [Analytics](/docs/workspace/analytics) open the list with a filter already set.
**Export CSV** downloads every conversation, not only the filtered ones. See [Exports](/docs/workspace/exports).
## Channels
| Badge | What it is |
| ----------------- | ---------------------------------------------------------------------- |
| **Chat** | The website chat widget. |
| **Phone** | A call to one of your agent's numbers. |
| **Email** | An email to the agent's address. |
| **Browser voice** | A voice call started in the browser, e.g. testing your agent by voice. |
## When a conversation ends
* **Phone and browser voice:** when the call ends. A browser voice call with no activity for 30 minutes is closed automatically.
* **Chat and email:** after 30 minutes with no new message. If the customer writes again later in the same thread, the conversation reopens. Any takeover is cleared when it reopens, so the agent answers first. If they ask for a person again, that is a new request.
## Summary, outcome, topic and sentiment
When a conversation ends, WireDesk reads the transcript and writes four things on it, usually within about fifteen minutes. Until then the conversation shows **Ended · summary on its way**. A live one shows **Live now**.
| Field | What it is |
| ------------- | ---------------------------------------------------------------------------------------- |
| **Summary** | One or two plain sentences: what they wanted and how it ended. |
| **Outcome** | One of the four below. |
| **Topic** | A two-to-four-word phrase, such as "appointment booking". |
| **Sentiment** | **Positive**, **Neutral** or **Negative**. This is the customer's mood, not the agent's. |
| Outcome | Meaning |
| ----------------- | ------------------------------------------------------------------- |
| **Resolved** | The agent answered what they came for. |
| **Handed to you** | A person took over: a live transfer, or a reply from the dashboard. |
| **Message taken** | The agent could not help and took a message for you. |
| **Abandoned** | They left before the agent could help, or never said anything. |
A conversation in which the customer never said anything is marked **Abandoned** straight away and gets no summary. For email, the email's subject stays in place of a summary. It appears as **Subject** in the conversation's details.
## A single conversation
The header shows the contact, the agent, and a sentence on where things stand, such as "Still going. New lines appear here as they arrive." or "They asked for a person and nobody has replied yet". The badges beside it show the channel, outcome or state, sentiment and handover state.
| Card | What it holds |
| ---------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **What happened** | The summary and topic (not shown for email). |
| **Transcript** | Every message in order. A live conversation updates every few seconds while the tab is open. |
| **Reply as owner** | The reply box and takeover buttons, on chat and email only. See below. |
| **Details** | **Started**, **Duration** (calls) or **Last activity** (chat and email), **Ended**, **Agent**, and, where they apply, **Handled by**, **Waited for a person** and **Subject**. |
| **What it looked up** | Each search the agent made of its knowledge base: the question, and how many passages it found and from which sources. |
| **Teach** *agent name* | Write the answer the agent should have given, then **Add to knowledge**. It is saved to the agent's knowledge base straight away, filed under the customer's question. See [Fix a wrong answer](/docs/knowledge/fix-a-wrong-answer). |
| **This person** | How many conversations this contact has had with your agents, and **Full history**, which opens their [contact page](/docs/workspace/contacts). |
| **Lead captured** | The lead the agent saved in this conversation, with a link to **All leads →**. |
| **Message for you** | A message the agent took, whether it has been read, and where to reach them. |
### Who wrote what
In the transcript, the customer is labelled **Visitor** in chat, **Customer** in email and **Caller** on calls. The AI is **Agent**. A teammate's reply is shown under that teammate's own name, so a conversation several people worked on reads clearly.
## Take over a chat or email
You can step in on any chat or email conversation from one of your agents. Phone calls are handed to a person differently: the agent transfers the caller live to your transfer number. See [Hand calls to you](/docs/actions/hand-calls-to-you).
Conversations where a visitor asked for a person show a **Needs human** badge, and chats waiting for someone also appear under **Needs attention** on the [dashboard](/docs/workspace/dashboard).
Click **I'll take this one** (when someone asked for a person) or **Take over from the agent** (when nobody asked). This pauses the agent and tells the visitor you've joined, before you type anything. It also shows your teammates that you have it. Sending a reply does the same thing.
Type in the box under **Reply as owner** and send. In chat, the visitor sees your reply in the widget, with your name. In email, it is sent from the agent's address in the same email thread. Replies can be up to 2,000 characters.
Click **Hand back to the agent** when you are done, and the agent answers again.
While someone has taken over, the badge reads **Human takeover**. Teammates see "*Name* is on it" next to the button.
A takeover hands itself back after 30 minutes with no reply from the person holding it, so a closed tab never leaves a customer talking to a silent agent.
### When a visitor asks for a person
If the agent's **Request a human** action is on and it is within your business hours, the agent can flag a conversation for a person. See [Request a human](/docs/actions/request-a-human) and [Handover](/docs/concepts/handover).
1. The conversation is marked **Needs human**, and you get a [notification](/docs/workspace/notifications). The agent keeps answering in the meantime, because a request does not pause it.
2. If nobody joins within 10 minutes, the request times out and the badge changes to **No answer**. In chat, the agent tells the visitor: "Nobody's free to join right now, sorry. Leave a way to reach you and someone will follow up." Then it carries on.
3. If someone joins, the time the visitor waited is recorded on the conversation as **Waited for a person** and feeds the **Answered by a person** figure in [Analytics](/docs/workspace/analytics).
If the visitor asks again while a request is still waiting, the notification says how many times they have asked. You are not notified more than once every five minutes for the same conversation.
## Related
* [Handover](/docs/concepts/handover)
* [Inbox](/docs/workspace/inbox)
* [Contacts](/docs/workspace/contacts)
* [Fix a wrong answer](/docs/knowledge/fix-a-wrong-answer)