Skip to main content
Every section before this one is about building the agent: giving it the right context, testing it, monitoring it, setting it up to improve itself, and shipping it into your product, Claude or Slack. This section is about the result: the analytics agent your users actually use. By the time it reaches them, it knows your business, every version has passed your tests, and its answers are monitored and improving. Your users see none of that machinery. They ask a question in plain English and get a trusted answer, and then they do something with it:
  • Keep it: save an answer to Favorites and come back to it.
  • Share it: send one answer, or a whole conversation, as a link.
  • Build with it: arrange saved answers into a canvas they can share.
  • Act on it: set up rules that send an alert or a report, or make a change in your product through your own API.
  • Rate it: tell you, with a thumbs up or down, whether the answer was right.
What they do feeds back to you: their ratings appear in Chat History beside the automated evaluations, and Self-improvement reviews the conversations that went wrong and suggests fixes. That closes the loop back to the building sections.
Home in User mode: a question asked in natural language and the chart the agent produced, with Edit chart manually and Add to favorites beneath it

The loop

Giving Feedback

Thumbs up and thumbs down on any answer, with a structured feedback form when something was wrong.

Sharing an Analysis

Turn one answer, or a whole conversation, into a link anyone can open.

Favorites

Save a chart or query, come back to it later, and pick the conversation back up where it left off.

Canvas

Arrange favorites into a live or snapshot board and share it privately, org-wide, or publicly.

Rules & Actions

Send an alert or a report, or call your own API, when the data on a canvas says so.

API Connectors

Let the agent and rules make changes in your product. Changes from chat wait for the user to confirm.

Where these features are available

Everything in this section works in the places an agent runs:
  • The workspace Playground — while you’re building and testing, so you can try the end-user experience yourself before shipping it.
  • Embedded in your product — the embedded agent exposes the same feedback, sharing, favorites and canvas surfaces to your users, scoped to their tenant.
  • Claude — the Upsolve Analytics connector asks questions, saves favorites and builds canvases from a conversation in Claude. What it saves is the same as what you save in the Hub.
  • Shared links — a shared analysis or a public canvas opens for someone who has no Upsolve account at all.
Favorites and Canvas are currently marked Beta in the product. They are functional and in production use, but the surface is still moving — expect refinements.

Why it matters for the builder

Each of these surfaces is also a signal:
  • Feedback lands in the Feedback column of Chat History, next to the automated evaluations, so you can compare what users think with what the evaluators think.
  • A favorited chart is a user telling you which answers are worth keeping — good raw material for a test case or golden asset.
  • A conversation that gets shared or forked repeatedly is a question your agent should be answering well every time.