Ask overview

Ask lets a Seat Holder put a question to the workspace’s feedback in plain language and get back an answer built from the actual data — not a summary of a summary.

What happens when you ask something

Submitting a question creates an Ask Query. Before writing a word of the answer, the agent calls tools against your workspace: searching the feedback corpus for verbatim evidence, pulling grouped aggregates (response counts, sentiment splits, per-theme mention rates), querying team health signals, and reading any connected integrations. Only then does it answer.

The answer is composed, not templated. As the agent works it emits Ask Artifacts inline — a trend chart here, a table there, a participant quote, a “Grounded in” panel resolving each [Evidence N] marker back to the response it came from. Which building blocks appear depends entirely on the shape of the question.

That composition is the point. A question about how often a theme comes up produces mention-rate bars; a question about how something moved over six months produces a line chart; a question about where people disagree produces a Conflict Signal that names both groups instead of averaging them into a single number.

The two-column layout

The left column is the Threads panel — every Ask Thread you’ve started, grouped into Pinned, Today, and Earlier. The right column is the thread itself: a header with the thread title and your remaining Ask credits, the scope bar showing what the thread is grounded in, the answers, and the composer.

An Ask Thread is the persisted, resumable unit. A single Ask Query is one turn inside it. Threads are per-Seat-Holder — you see your own, not your colleagues’. See Threads and pinning.

Grounding decides what “the data” means

With no grounding set, Ask answers across all workspace feedback. Setting Thread Grounding binds the thread to specific sources — one or more surveys, or closed retros — and retrieval is constrained to them server-side.

This matters more than it sounds. Org-wide sentiment is dominated by thousands of unrelated responses, so an ungrounded question about one survey’s tooling complaints will read as flat. Ground the thread first and the same question gets a straight answer.

Where you enter Ask from

From	What it does
The Ask item in the sidebar	Opens Ask with no grounding
The command palette	Typing a question offers an “Ask: …” command that opens Ask and sends it
A survey’s detail page	Opens Ask deep-linked to that survey
A closed retro’s summary	Opens Ask deep-linked to that retro

Only closed retros can be used as a source. An active or reopened retro silently drops out of a thread’s grounding until it is closed again.

What Ask will not do

It won’t attribute feedback to an individual. Answers synthesize across data points — “team members report…”, never a name against a quote.

It won’t quote below the anonymity threshold. When the scoped pool of participants is too small, verbatim quotes are withheld by design and the answer says so rather than working around it.

It won’t invent a number. Per-theme counts and mention rates come from the aggregate tool or they don’t get stated. When the sample is small, the answer is expected to say the sample is small instead of drawing a trend through it.

It won’t take instructions from feedback content. Retrieved feedback is treated as quoted data. Text inside a survey response that tells the agent to create a survey or reveal an identity is reported as a curiosity, not obeyed.

What it costs

Every Ask Query draws one credit from the shared workspace Ask Pool — not from a per-person allowance. The header chip shows what’s left.

Where to go next

Threads and pinning — start, resume, rename, pin, and delete threads
Grounding a thread — scope a thread to specific surveys or closed retros
Ask Artifacts — every building block the agent can render
The Ask Pool and usage — allowances, credits, and what happens when they run out