one workspace, every level of the same question.

every conversation arrives already read. this is the room you read it back in.

start from any number. it opens into the conversations underneath it.

take the tour request access

set the scope once. look at it from every level.

pick a cohort, a topic, a task, an intent or a signal. every surface reads from that scope until you change it.

your overview is written in sentences, not tiles.

it names the task that moved, the intent underneath it, and the cohort carrying the change. every clause links to the conversations it came from.

every number carries its previous value, so the brief reads as a direction and not just a level.

search by meaning. replay with the hard parts already marked.

describe what went wrong in plain english. every matching task comes back, however it was phrased. open one and the flagged turns are already waiting.

  • friction markers pinned to the exact turn
  • annotated replays you can share
  • plain english in, conversations out

a person, written as a paragraph.

every person's page opens with prose. what they keep asking the agent to do, what finished, and which signal keeps turning up. their full history sits underneath it.

three filters, and the step that fails is on screen.

open a topic, pick the task, keep the failed attempts. the task splits into the intents inside it, and every bar opens into the conversations behind it.

topic · product questions task · summarise what my team shipped task status · failed 864 conversations
read it528 stopped · 22.9%
write the summary244 stopped · 13.7%
find the source92 stopped · 3.8%

the filters are the query. what you picked stays readable at the top, so anyone can see how the number was made.

the alert arrives while the conversation is still open.

a monitor is built from the view you are already looking at. it fires when a task starts failing, an intent spikes, or a cohort drifts away from the rest.

it goes to slack, email or a webhook. it carries the scope, the size of the move, and a link into the matching conversations.

the same task does not succeed everywhere.

group people by plan, region, or how often they come back. the task stays fixed and the group varies, so a difference is a difference.

enterprise, EMEA 986 users 52% read it documents came back empty
EMEA 58% read it documents came back empty
returning users 61% read it repeated request
enterprise 66% write the summary user correction
self-serve 71% write the summary user correction, then recovery

the baseline is 64%. across these cohorts the spread is 19 points. the highlighted row is the one worth opening, and 986 users sit inside it.

connect the agent. see where the task stops.

one workspace for the conversation, the pattern around it, and the people on both ends of it.

no email, no call. the playground is the live workspace.