Keypup's AI Agent is a data analyst that works on your real engineering data. It builds the dashboard you need, explains what the numbers say, finds the cause behind them and recommends actions you can track.
Four capabilities that form one loop: from the question you ask to the action you take, and back to the dashboard that proves it worked.
Describe what you want to understand. The agent creates the dashboard: it finds relevant templates in the library and your collections, configures each insight, writes custom formulas, sets drill-downs and dashboard filters, and adds notes explaining what is measured.
Click Analyze on any dashboard. The agent reads the results of every insight and tells you what they say: trends, outliers, how metrics move together, and which numbers deserve attention.
It goes past the average: the workflow stage where work stalls, the stale issues inflating a cycle time, the review load concentrated on one person, the untracked work slowing the roadmap.
It proposes concrete actions, such as WIP limits, smaller pull requests or a better spread of reviews. Ask, and it builds the dashboard that tracks whether they worked over the next sprints.
The tracking dashboard becomes the next question: the loop starts again, sprint after sprint.
The agent is available wherever you work with data: on an empty dashboard, beside any dashboard, and inside the insight editor.
Three diagnoses the agent produced on dashboards it built in our demo workspace. Each starts with a question and ends with actions you can track.
"Why is our cycle time far above our 6-month benchmark?"
Average time in progress is close to five times the 6-month benchmark. A handful of stale issues, open for thousands of hours, inflate the average, and cycle time spikes for very large pull requests.
"Where is our flow suffering, and is anyone at risk of burnout?"
Large pull requests drag flow efficiency and review time. Reviews are concentrated on one senior engineer, and after-hours activity peaked during one month of the period.
"How much of our work never reaches the roadmap?"
A large share of commits is not linked to any Jira issue. That untracked work adds measurable time to cycle time and makes team capacity invisible to stakeholders.
Good analysis needs context, and good decisions need oversight. The agent is built for both.
Connect Claude, ChatGPT, Cursor, VS Code or Kiro to the Keypup MCP server. Your assistant gets Keypup's natural-language query engine, your datasets and your team's AI Instructions, and can analyze and diagnose right where you already work.
get_company_ai_instructions generate_dataset_query query_dataset query_dataset Analysis. Lead time for changes rose from 2.1 to 3.4 days. Coding time is stable; the increase is in review.
Diagnosis. Time to first review doubled on the payments repositories, where two reviewers handled 70% of reviews.
Recommendation. Add reviewers to the payments rotation and set a 24-hour first-review target.
Illustrative conversation
hq.keypup.io/mcp, in one click with OAuth or with an API token.The MCP server is read-only: your assistant can query and analyze, but cannot change your Keypup dashboards or your source tools.
Set up the MCP serverConnect your tools, open an empty dashboard and tell the agent what you want to understand.