# Threads page URL: /docs/cloud/dashboard/threads Search every conversation, archived ones included, filter it by behaviour or analysis, and inspect its transcript, runs and interactions. > For AI agents: a documentation index is available at [llms.txt](/llms.txt). Use `.md` for canonical markdown pages; `.mdx` is kept as a backwards-compatible alias on supported URL paths. Threads is the conversation list. It starts with every thread in the selected range, archived ones included, then lets you narrow the list by what people discussed, how they reacted, how they were reviewed and what the project's analysis found. ![Threads page on the demo project](/_next/static/immutable/media/threads.0aedna2r_m7qk.webp) ## Search the conversation list Archived threads are listed too, with an Archived mark beside the title: an end user archiving a thread is housekeeping on their side, and the thread's runs, cost and ratings still belong to the project, so a figure that links here opens exactly that many rows. The volume strip shows threads in each time bucket. Select a bar to narrow the list to that bucket. Search matches a thread title or the full text of stored message content. The table has **Title**, **Topic**, **User**, **Tokens**, **Cost** and **Updated** columns. **Topic** appears once conversations are classified. A thread's thumbs up and thumbs down counts and its latest review verdict appear as marks beside the title, only on the threads that have them. **Tokens** and **Cost** can be sorted. The table loads 50 rows at a time. [Export](/docs/cloud/dashboard/exports) appears on plans that include exports. ## Filter by conversation and behaviour The filter bar always shows **Feedback**, **Review**, **Satisfaction**, **Signal**, and **Ended on the user's message**. Other facets appear only when they have options in the selected range. Facet values describe the threads in that range, archived ones included. | Facet | Values and reading | | ------------------------------- | ----------------------------------------------------------------------------------------------- | | **Topic** | Analysed nonjunk topics, ordered by count. | | **Task** | Analysed tasks. | | **Satisfaction** | `satisfied`, `mixed`, `dissatisfied` or `none`, using the rule below. | | **Feedback** | `feedback=negative`, `feedback=positive` or `feedback=any`. | | **Review** | `review=unreviewed`, `review=reviewed`, `review=good`, `review=needs_work` or `review=unclear`. | | **Signal** | A multiselect of the 12 interaction and analysis signals below. | | **Environment** | Run environments, selecting threads with a run in that environment. | | **Sentiment** | `positive`, `neutral` or `negative`. | | **Resolved** | Resolved or unresolved analysis results. | | **Language** | The 12 most common analysed languages. | | **Junk** | **Hide junk** selects nonjunk threads. **Only junk** selects junk threads. | | **Question** | Analysed questions, up to 50 options. | | **Ended on the user's message** | Threads whose last message by creation time has the `user` role. | | Signal value | Label | Reading | | ------------- | --------------------- | ------------------------------------------------- | | `stopped` | Stopped | A `run_stopped` event. | | `regenerated` | Regenerated | A `message_regenerated` event. | | `copied` | Copied | A `message_copied` event. | | `edited` | Edited and resent | A `message_edited` event. | | `attachment` | Attached a file | An `attachment_added` event. | | `suggestion` | Clicked a suggestion | A `suggestion_clicked` event. | | `branch` | Switched branch | A `branch_switched` event. | | `approved` | Approved a tool | A `tool_approved` event. | | `rejected` | Rejected a tool | A `tool_rejected` event. | | `error` | Saw an error | An `error_shown` event. | | `no_reply` | No reply | The analysis found no assistant messages. | | `loop` | Repeated or rephrased | The analysis found an exact repeat or a rephrase. | Satisfaction reads the thread's ratings first and its behaviour second. A thumbs up on any message makes the positive side true, and a thumbs down on any message makes the negative side true. A rating outranks behaviour: a copy or second turn is positive only on a thread with no thumbs down, and a stop, regeneration or edit is negative only on a thread with no thumbs up. A thread with both ratings is mixed. ``` positive = threads.thumbs_up OR (NOT threads.thumbs_down AND (threads.copied OR threads.run_count >= 2)) negative = threads.thumbs_down OR (NOT threads.thumbs_up AND (threads.stopped OR threads.regenerated OR threads.edited)) satisfied = positive AND NOT negative mixed = positive AND negative dissatisfied = negative AND NOT positive none = NOT positive AND NOT negative ``` **Satisfied** has only a positive side, **Dissatisfied** only a negative side, **Mixed** both, and **No signal** neither. A thumbs rating from an end user is never a review. ## Review threads **Review** is a second view next to **List**. A thread counts as reviewed once any person has reviewed it, and its verdict is the latest review recorded on it, whoever wrote it. The view uses the same filters as the list and reports the reviewed count out of the total threads in the current selection, counts for each verdict, agreement between that latest **Good** or **Needs work** verdict and the end user's thumbs on threads that have both, and the latest notes. These figures cover the threads selected by the filters, not the whole project. **Start reviewing** opens the first unreviewed thread in the current filter and keeps the walk on unreviewed threads. Starting with **Feedback** set to **Thumbs down** is useful when reviewing dissatisfaction. ## Inspect one thread Select a row to open its detail page. ![Thread detail page on the demo project](/_next/static/immutable/media/thread.1shsgq4goqf8y.webp) The top tiles are **Messages**, **Runs**, **Cost** and **Tokens**. The main column is **Conversation**, drawn by the dashboard or, once the project sets a Renderer URL, by your own app; [Conversation views](/docs/cloud/dashboard/conversation-views) says what each draws. The right rail holds the rest of the thread's reports and analysis. | Place | What it shows | | ---------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Conversation** main column | The rendered transcript. End user thumbs and their comments appear under the message they were given on. | | **Details** tab | The details ledger, Satisfaction, and Analysis. Analysis includes Topic, Task, Phrasing, Sentiment, Resolved, Language, Turns, Repeats, Unanswered question, and Classified. | | **Turns** tab | One row per question with the run that answered it. Open a run row to see its spans and **Annotate** action. | | **Activity** tab | The engagement events for the thread. | | **Review** tab | The verdict form and other reviewers' reviews. | | **Raw** tab | The stored messages as JSON. | The right rail has five tabs: **Details**, **Turns**, **Activity**, **Review** and **Raw**. It opens on **Details**; a thread opened from the **Review** view or from a list filtered by review state opens on **Review** instead. On the **Review** tab a person on the team records a verdict for the whole conversation: **Good**, **Needs work** or **Unclear**. They can add an optional note, then choose **Save** or **Save and next**. `g`, `n` and `u` choose the verdict. `Mod+Enter` saves and moves on. `j` and `k` move to the next and previous thread. The header links back to the list the thread was opened from: **Back to threads**, **Back to user**, **Back to topic** or **Back to task**. It also includes **Exit review** while a review walk is active. **Previous** and **Next** stay within that list's range, filters and sort, including a user's threads or an Intelligence topic or task. The counter reads **N of M** outside review and **N left of M** in review. Each reviewer has one review per thread and can change it. The **Save** control reads **Update** once the viewer has a review. Other reviewers' reviews are listed below the form. The **Details** tab has **External id**, when the thread has one and with a copy control, then **Title**, which reads how the thread got its name or why it has none, then **Created**, **Last message** and **Runs**. **Environment** and **Release** show the distinct values across the thread's runs and link to [Runs](/docs/cloud/dashboard/runs) filtered to that value. **Tags** are badges that link to [Runs](/docs/cloud/dashboard/runs) filtered to the tag. The **Metadata** ledger appears only when the thread has metadata. | Title row | Meaning | | -------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------ | | *Written by gpt-5.6-luna on the first message in 1.4s* | The model, the trigger that titled the thread (*on the client's request*, *on the first message* or *by the hourly sweep*) and how long the call took. | | *Set by the client* | The title came from a rename or a thread update. | | *Being written* | An attempt is running. | | *Attempt 1 failed: Invalid API key. The hourly sweep retries.* | The last attempt failed with the provider's error; the sweep retries up to three times. | | *Gave up after 3 attempts: Invalid API key* | Every attempt failed. Rename the thread. | | *Waiting for the hourly sweep* | No attempt yet; the sweep picks the thread up. | | *No user text to title* | The thread has no user message with text; a message that holds only a file is not title input. | | *Titles are off for this project*, *Paused by the plan cap* | The feature is off, or the gateway allowance is used up; see [Thread titles](/docs/cloud/thread-titles). | | *No model configured for titles* | The project has no configured title model. | **Export as Markdown** appears on each assistant message and downloads only that message as `message-.md`. ## Troubleshooting | What you see | Why | What to do | | ---------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------ | | A thread is missing from the list. | The active range, search and selected facets limit its rows. | Clear the list filters and choose a range that includes the conversation. | | A facet is missing from the bar. | It is not one of the facets shown in every range, and it has no options in the selected range. | Widen the range or remove the filters that leave that facet with no options. | | Satisfaction reads *No signal*. | The thread has no thumbs rating and neither a copy or second run nor a stop, regeneration or edit. Its stored satisfaction value is `none`. | Read the **Activity** and **Turns** tabs to see the events and runs behind the rule. | | Analysis reads *Not analysed yet*. | No [Intelligence](/docs/cloud/intelligence) pass has classified the thread: analysis is off, the thread is newer than the last pass, or it was stored as junk. | Check **Settings › Intelligence** and its runs ledger, or queue **Run now**. |