Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.wittify.ai/llms.txt

Use this file to discover all available pages before exploring further.

Ask & Visualize is the page that lets you ask plain-language questions about your contact center and get charts, numbers, or raw rows back in seconds.

Page anatomy

RegionWhat you see
Filter barDate and hour controls, see Overview.
Sidebar (start edge)A list of past chat threads with search, New chat, and per-row delete.
Chat canvas (center)The main scrollable area. Question turns from you, answer turns from the system.
Composer (bottom)The text input plus Send button. The composer pins to the bottom of the page.

The composer

Pinned at the bottom of the chat canvas.
ElementWhat it shows
Text inputPlaceholder Ask anything about your calls… Capped at 500 characters. Arabic input renders right-to-left automatically.
Send buttonA brand-gradient pill with a send icon. Disabled when the input is empty or a turn is currently streaming.
Stop buttonReplaces Send while a turn is streaming. Click to cancel the in-flight answer.
The Send button has extra padding on the end edge so the floating chat widget launcher in the bottom-end corner does not cover it.

The thread sidebar

A list of every chat thread you have started in this project.
ElementWhat it shows
History headerTitle at the top of the sidebar.
Search boxPlaceholder Search history… Filters the thread list by title.
+ New chat buttonOpens a fresh thread.
Thread rowEach row shows the thread title (the first question) and a small chevron.
Active filter pillBelow each thread, a small chip shows the date range that thread was scoped to.
DeleteClick the small trash icon on a row. The first click shows Click again to confirm; the second click deletes the thread.
Empty stateNo chat history yet.

The chat canvas

Scrolls vertically. Older turns at the top, newest at the bottom. Each turn carries:
ElementWhat it shows
AvatarRound tile, brand gradient for system turns, your initials for your turns.
BodyThe text or rendered chart.
Per-answer footerSmall action buttons (described below).

Streaming and error states

StateWhat you see
Question sentThe question turn appears immediately, the answer turn shows a spinner with Thinking….
StreamingThe answer renders progressively as the model thinks and the SQL runs.
FailedThe answer carries the message Something went wrong. Try again. in a small red strip.

Suggestions

When the canvas is empty, the page surfaces six suggestion chips you can click to drop a question into the composer.
Suggestion
Show me total calls for the last 7 days
Top 10 intents across the dataset
Sentiment distribution this month
What is the busiest hour of the day
Show an activity heatmap by day and hour
List the most recent negative sessions
The empty state title reads Ask anything about your contact center. with a body line I can summarize call volume, surface top intents, plot sentiment, and show raw session rows. Try one of the suggestions below or type your own question.

What the system can return

The system picks the best renderer for your question.
RendererWhen it shows up
Big numberSingle-value answers (e.g. “Total calls last 7 days” → one big number with a comparison delta below).
Pie chartDistribution questions (e.g. “Sentiment distribution this month”). Donut shape with inside-positioned percent labels. Slices under 4% hide their label. The legend is truncated at 28 characters and scrolls.
Bar chartRanking questions (e.g. “Top 10 intents”). Vertical for time series, horizontal for ranked categories. Each bar carries a small data label at the tip.
Line chartTime series (e.g. “Calls per day this month”). The x axis is dates, the y axis uses thousands separators.
HeatmapCross-tab questions (e.g. “Activity by day and hour”). Same shape as the Activity Heatmap on Overview.
TableRow-level questions (e.g. “List the most recent negative sessions”). Up to a fixed page cap, with a row count and a Truncated for display indicator if more rows exist.
ActionWhen it shows
PNGOn chart answers. Downloads the chart.
CSVOn table answers. Downloads the data.
rows countOn table answers. Reads rows.
Truncated for displayOn table answers when the result was capped.

Common questions

The system picks the renderer based on the question’s shape (a count yields a big number, a distribution yields a pie, a ranking yields a bar). Re-phrase the question to nudge it (for example, “List the rows” forces a table, “Show me the share” forces a pie).
Two common causes. First, the question may be ambiguous (for example, “show me last week” could mean any of the past 7 days, or the previous Monday-to-Sunday week). Be explicit: Last 7 days or the most recent calendar week starting Monday. Second, the date filter at the top of the page applies, so a question like “Show me total calls” returns the count inside the active filter window, not the project’s all-time count.
The system mirrors your question’s language. If you got an English answer for an Arabic question, ask the question again in Arabic so the system has a clean signal.
No. Every query is read-only. Only SELECT and WITH queries are accepted, and destructive operations are rejected at the server. Sensitive columns are also deny-listed.
Yes. Click PNG on a chart answer or CSV on a table answer. Both are downloaded to your machine, you can attach them to chat or email.
Use the per-row trash icon to delete chats you do not need. The first click asks for confirmation, the second deletes. There is no undo.
Yes. The system keeps the thread context, follow-ups like “And by hour?” or “Now show only the negatives” work in plain language.
The composer’s end-edge padding clears the launcher, the Send button stays clickable. If your screen is unusually narrow and the launcher overlaps anyway, scroll the page slightly, the launcher slides out of the way.

Where to go next

Sentiment Analysis

Curated chart for sentiment distribution.

Intent Analysis

Curated chart for top 10 intents.

Volume Forecasting

Plan ahead with a model-based projection.

TimeLens AI

Streamed narrative report instead of a one-shot answer.