AI agents

Agents are the core of everything in Lightdash. They leverage your team’s knowledge and metrics to provide consistent answers, then build those components into a self-improving context layer, so you can focus on putting your data to work. Whether helping leadership monitor KPIs, data scientists forecast growth, or building data apps to empower your sales team with a custom GTM tool, Lightdash agents weave trustworthy data into your daily work.

​

Ask more questions

Set up an agent for your team, then start asking questions. You can manage as many threads as you need, and agents always have the right context. Compare freely across segments and time periods, get hands on exploring the charts and tables that come back, then zoom back out to the agent for more iteration. Because agents are integrated everywhere in Lightdash, you’ve got a full analytics platform to dig in with (unlike hooking Claude up to your database). When you find something useful, just ask the agent to build whatever you envision. You can go from spotting a trend to deep causal analysis, then building those insights into a new team dashboard, all in a single thread.

​

Get trustworthy answers

Traditional dashboards decay into sprawl the more a team uses them. Agents get better, because every question tests what your project knows. Metric questions run on the definitions and joins in your semantic layer, so agents start from the foundations your team has already built and keep raising the bar from there. When an agent picks the wrong table or misreads a term, Issues catches it in the background, traces it to a root cause, and drafts the fix: a missing metric, a clearer description, or a note in your shared project context. Once your team merges it, every agent that reads that context answers better. Autopilot keeps your content healthy on its own schedule, repairing broken charts, retiring stale ones, and drafting the charts people keep asking for. Evaluation suites built from real threads catch regressions before your team ever sees them.

​

Build your agentic data platform

Once teams rely on an agent, it stops being a one-off conversation and becomes part of how they work. Shape each agent around the team it serves. Give a finance agent your revenue recognition policy as a knowledge document, teach it how your team runs a monthly variance analysis with a skill, and point it at verified answers for the questions it should answer the same way every time. Set its data access and permissions to match the job, and you can hand it more ambitious work, from building dashboards to opening pull requests. With agents as code, all of that configuration, along with its skills and evaluation suites, lives in Git, goes through review, and promotes from a preview project to production. Conversations can change the project too. Ask an agent to add the metric it couldn’t find, and AI writeback edits your dbt or Lightdash YAML, validates it, and opens a pull request for your team to merge. The AI coding agent does the same for other repositories your organization connects, so a question in chat can end as a reviewed change wherever the fix belongs.

​

Integrate everything

The work doesn’t have to stay in Lightdash. Connect an agent to a Slack channel so a sales team can check pipeline numbers in the thread where they’re already discussing a deal, or embed an agent in your product so your customers can ask about their own data. When an analysis is something people should see every week, turn it into a recurring delivery from the same conversation. MCP works in both directions. Connect Claude, ChatGPT, Cursor, or your own agents to the Lightdash MCP server, and they get the same metrics, verified content, and agent context your team uses in Lightdash. They can build charts and dashboards or open a writeback pull request without leaving the tool they’re in. In the other direction, external MCP servers let your Lightdash agents act on what they find, like filing a Linear issue when a metric drifts or pulling a spec from Notion before building the dashboard it describes.