> For the full page index, fetch https://echonova.ai/llms.txt # Find and read a conversation Open the Conversations tab, narrow the list to the call or chat you want, and read what the agent and the caller said. Every call and every chat an agent handles is kept as one conversation. The **Conversations** tab on the agent lists them newest first and opens each one on its transcript, with the numbers measured while it ran. Use it to check what the agent said, what the conversation cost and how it closed. ## What you need - Membership of the workspace that owns the agent. Any member of the workspace can open the tab. Conversations are read with your portal sign-in, not with an API key. - An agent that has handled at least one conversation. See [What is an agent](/docs/agents/what-is-an-agent) for what a conversation is. ## Steps ### Step 1: Open the Conversations tab Open the agent, then the **Conversations** tab in the row beside **Overview** and **Persona**. Three cards sit above the list. | Card | What it shows | | ----------------- | ------------------------------------------------------------------------------------------------------------------------------- | | **Conversations** | How many conversations match the filters you have set right now. It follows the filters; the other two do not | | **Unique users** | Distinct contacts over the last 30 days, marked `30d` | | **Resolved** | The share of conversations that closed without failing over the last 30 days, marked `30d`, shown as a dash when nothing closed | ### Step 2: Narrow the list The toolbar carries one search box and two filter menus. **Search for conversation...** matches the customer name and the contact id. It is case insensitive, it does not look inside the transcript, and it runs 300 milliseconds after your last keystroke. **Type** offers **All**, **Voice** and **Chat**. **Chat** means every text conversation, whichever channel carried it. **Status** offers **All**, **Open**, **Resolved** and **Failed**. A conversation has one of four status values, and the three chips group them: | Chip | Status value | Means | | ------------ | ---------------------------- | ---------------------------------------------------------------------- | | **Open** | `active` | Still running. The agent may still answer on it | | **Resolved** | `ended` and `out_of_credits` | Closed. A conversation cut off by an exhausted balance also lands here | | **Failed** | `failed` | Closed by an error | Changing a search term or a filter sends you back to the first page. ### Step 3: Open one conversation Select a row, or open its actions menu and choose **View details**. Each row carries five things: - The customer, as a name when one is known. Voice, SMS and Twilio rows fall back to the contact id. Everything else falls back to a channel word such as **Web visitor**, **Slack conversation**, **WhatsApp contact** or **Ticket requester**, followed by a `#` and the last four characters of the contact id. - The last message, when the conversation has one. - The channel it arrived on. - The date it started, with the time under it. - Its status chip, red when the status is **Failed**. A phone icon marks a voice conversation and a message icon marks a text one. ### Step 4: Read the transcript Opening a conversation loads the newest 50 messages. The caller's turns sit on the left and the agent's on the right, with a date pill wherever the calendar day changes. **Load earlier messages** pulls the previous 50, and keeps appearing while there are more. Text in square brackets, such as `[laughs]`, is a delivery cue the agent wrote for its voice rather than something anyone said. The portal prints those in red. An agent message that carried a measured model time shows it beside the timestamp as `LLMS N Sec`. ### Step 5: Read the inspector The panel on the right opens on **Conversation Overview**. | Part | What it holds | | --------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------ | | Channel heading | The channel this conversation arrived on | | **Summary** | The agent's wrap-up of the conversation, when it wrote one, plus the tags on this conversation and the **Add Tag** control | | **Metadata** | Three measured cards, a **Copy ID** button, and the facts list below it | | **LLM cost** | One point per agent message that recorded a model credit share, plotted across the conversation. With nothing measured it reads `No data has been collected` | The three measured cards are **Avg. response time**, then **Turns** on a voice conversation or **Messages** on a text one, then **Call duration** on a voice conversation or **Duration / session length** on a text one. Duration is the wall clock between the start and the end, so it is blank while the conversation is still open. The facts list under them holds the connected account when the conversation came through one, the user, the conversation id, the date, the status, **Total credits**, and **Tools triggered**. **Tools triggered** counts the distinct tools this conversation used against the tools the agent has enabled, which is its switched-on built-in tools plus its selected workspace tools. ## Get it right Search reads the customer name and the contact id. Start from who the conversation was with, then narrow by type and status. **Recommended** - Paste the contact id or type the customer's name, then set **Type** and **Status**. - To find a call that failed yesterday, set **Type** to **Voice** and **Status** to **Failed**, then read the dates. **Less effective** - Typing a phrase from the transcript into the search box. - Scrolling the whole list with every filter on **All**. ## Worked example A customer says the agent stopped answering them on WhatsApp this morning. 1. Set **Type** to **Chat** and **Status** to **All**. 2. Paste the customer's phone number into the search box. The list narrows to that contact's conversations, newest first. 3. Open the top row. The status chip and the end of the transcript show how the conversation closed, and **Total credits** in **Metadata** shows what it cost. 4. Press **Copy ID** and keep the id in case you raise the case with support. ## What happens next The list keeps itself current while you are looking at it. A conversation that changes shows the change within a second. The conversation id in **Metadata** is the id to quote in a support request. **Copy ID** puts it on your clipboard. From here you can label the conversation. See [Tag a conversation](/docs/agents/conversation-tags). For what a conversation records turn by turn and why a summary is sometimes absent, see [Transcripts and summaries](/docs/agents/transcripts-and-summaries). ## Limits | Limit | Value | Applies to | | --------------- | --------------- | ----------------------------------------------------------- | | Rows per page | 8 | The agent's **Conversations** tab | | Search text | 100 characters | The search box | | Transcript page | 50 messages | The transcript, and each press of **Load earlier messages** | | Summary | 4000 characters | The **Summary** above the transcript | ## Troubleshooting ### Couldn't load conversations The list could not be fetched. The tab shows that title, the body **Check your connection and try again.** and a **Retry** button. Retry keeps the filters you had set. ### No conversations found. The filters matched nothing. Set **Type** and **Status** back to **All** and clear the search box before concluding the agent has no traffic. A conversation belongs to exactly one agent, so a conversation you remember may be on another one. ### Your search returns nothing although you can see the words in the transcript Search covers the customer name and the contact id, not the message text. Find the conversation by its contact, open it, and search the loaded transcript with your browser's find. ### A conversation reads Resolved although the call was cut off **Resolved** covers both `ended` and `out_of_credits`. A conversation that ran out of credits mid-call is closed, not failed, so it is counted as resolved here and in the resolution rate on the analytics pages. Open the conversation and read the end of the transcript to see how it closed. ### The customer shows as Visitor with four characters after it The channel supplied no name for that contact. A row shows a name only when the channel provides one; otherwise the channel word and the last four characters of the contact id stand in for it, so two rows with the same word are still different people. ## FAQ ### Can I delete a conversation? Yes. Open it and press **Delete**, then confirm. Deletion is final. A conversation that is still open, or ended less than 15 minutes ago, cannot be deleted yet. See [Data retention](/docs/agents/data-retention). ### Can I export a transcript? **Copy ID** copies the conversation id. For a copy of a transcript, write to privacy@echonova.ai with that id. ### Does an open conversation count towards anything? It counts as a conversation the moment it opens, which is why the counts can include calls that are still running. It contributes no duration and no resolution until it closes. ### Do playground sessions appear here? Yes. A **Chat** test appears as a web chat conversation and a **Call** test as a web voice one, each with its transcript and credits. Nothing marks them as tests. ### Can I filter by tag? No. The list filters on status, on type and on the connected account, and the search box reads names and contact ids. Tags are shown once a conversation is open. ## Related - [Transcripts and summaries](/docs/agents/transcripts-and-summaries): what a conversation records turn by turn - [Tag a conversation](/docs/agents/conversation-tags): label the conversation you just read - [Agent analytics](/docs/agents/analytics): the same conversations as numbers - [The agent overview tab](/docs/agents/agent-overview-page): the five cards above the recent conversations - [What is an agent](/docs/agents/what-is-an-agent): what a conversation is