Looky was built so your agent could use it
Everything in Looky is plain text — Malloy models, YAML visualizations and dashboards — published through a CLI and validated before anything goes live. That is exactly the surface AI coding agents are good at. So instead of embedding a chatbot, Looky gives your own agent everything it needs to do the work, while you stay in charge.
- Your agent, your account. You bring the agent you already use and trust. Looky ships no AI, runs no AI, and never sees your AI credentials.
- You approve what ships. Content is validated before publishing, and publishing always requires your explicit yes.
- Viewers are untouched. Your audience keeps seeing dashboards in the browser, exactly as before — the agentic workflow is for the people who build.
Pick your agent
Available now
Claude Code
A ready-made workbench folder: download it, open it with Claude Code, and the agent sets everything up — from installing the tools to publishing your first change — teaching you as it goes. Requires a paid Claude plan.
Available now · free plan
Cursor
The same workbench, packaged for Cursor — and reachable on its free plan. Built to live with usage limits: it saves progress as it goes and can hand the commands back to you when the quota runs out.
More agents next (Antigravity and others): same workbench idea, one template each, tuned to how that agent works.
Already have your own setup?
Any agent that can read documentation and run a CLI can work with Looky today. Point it at the machine-readable docs — looky.studio/llms.txt — give it a folder linked with looky login, and it has everything: the content contracts, the build order, and looky validate as its safety net. The workbenches above just make that setup effortless.
What you need in every case: access to your Looky instance (an invitation from your account's owner or a developer role) and an agent of your choice. Data source credentials remain your team's responsibility — agents author content; your developer or owner deploys source settings.