Pilea answers research questions about your customer feedback. You ask a question in plain language - “What are the most common complaints from enterprise customers?” - and Pilea searches your mentions, backlog, surveys, and uploaded documents to write an answer with evidence.

Every workspace has one agent, called Pilea. It carries a set of skills (reusable instructions), and each skill can run on a schedule. This article covers how a single question or run works. For skills and schedules, see Agents, skills, and pulses.

What Pilea reads

When you ask a question or a skill runs, Pilea uses a series of tool calls to search and synthesise your workspace data. It reads:

Mentions - individual feedback entries from all your integrations and survey responses
Backlog items - issues and feature requests Pilea has identified across your feedback
Survey responses - answers and summaries from any surveys you’ve run
Documents - files, transcripts and spreadsheets you’ve uploaded under Context, plus cowrite documents

Pilea does not read your source code, billing data, or any data outside your Pilea workspace. It does not store your question or the generated result outside of Pilea.

Two ways to get answers

Skills (scheduled): A skill is a fixed instruction that Pilea runs on a schedule you set. The result is posted to the Activity feed and, if you choose, to a Slack channel, email, or both. This is useful for recurring questions like “What feedback came in this week?” or “Which backlog items are growing in mentions?”

Ad-hoc questions (chat): Click Chat in the top right and ask. Pilea answers straight away in the conversation. Use @ in the prompt to reference specific backlog items, customers, tags, themes, surveys, or uploaded documents and focus the search, or type / to run a skill on demand. See Chat and cowrite.

Reports across a whole period

When you ask for a report, summary or trend analysis of feedback over a date range, Pilea first builds a full-coverage basis: exact totals and breakdowns over every matching item, and a theme digest produced by summarising all of them in batches. It states the number of items it covered in the report. If part of the set could not be processed, it says so rather than silently summarising a subset.

Understanding run status

Each update in the Activity feed shows a status as Pilea works:

Status	What it means
Pending	The run is queued and will start shortly
Generating	Pilea is actively searching and writing the result
Completed	The update is ready to read
Failed	Pilea could not find enough data. The row shows “Insufficient data”.

Most runs move from Pending to Generating to Completed in under a minute. If your workspace has a large number of mentions, or the question spans a long period, it may take longer.

ℹ️ Note: A “Failed” run with “Insufficient data” usually means the question was too narrow for the data available - for example, asking about a customer segment with no feedback yet. Try broadening the question or checking that the relevant integrations are connected.

What’s inside a completed update

Open any update to read the full result. It includes a written narrative answering the question, with references to the mentions and backlog items used. Updates can include charts.

Below each section you can ask a follow-up question; the answer is researched and added to the same update. You can also Download PDF and Save for later from the update’s toolbar.

Who can see an update

Runs produced by a scheduled skill are visible to everyone in the workspace through the Activity feed. Updates that are private to you show Only me in the feed.

Agents, skills, and pulses - how to set up a skill, its schedule and where it delivers
Chat and cowrite - asking ad-hoc questions and referencing workspace items

How the Insights Agent works