An analytics platform
Pull data from any source
Files, databases, Python, email — each source becomes a node. No connectors to maintain. No per-source pricing.
A data lakehouse, included
Every source lands in DuckLake — versioned, cataloged, queryable from one place. No warehouse to provision.
Combine sources into one view
Join across sources with reactive SQL and Python. Change anything upstream, everything downstream re-runs. No cron jobs, no orchestrator to babysit.
Explore the unified view
Open a notebook. Ask the built-in agent for charts, transforms, or insights. Connect Claude, Cursor, or any MCP client to work directly in your graph.
Share the result
Build a dashboard from the same nodes you transformed and analysed. Cross-filtering works without wiring. Share a link — stakeholders see live data, not last Tuesday's screenshot.
How it works
It's a spreadsheet for data work
Every SQL query, Python script, and chart is a node on the graph, and every node knows what it depends on. Change a query upstream, everything downstream re-runs. New data lands, dashboards refresh on their own. And because the connections are explicit, you can trace any number back to its source.
Pipeline, notebook, dashboard. Same graph.
The same nodes show up as a pipeline (to build), a notebook (to explore), or a dashboard (to share). Different surfaces for different audiences, same source of truth underneath. Switch any time without rebuilding.
Interactive without wiring
Clicks and parameter changes are just inputs to the graph. Click a bar, every chart re-queries. Move a slider, every downstream query re-runs. The interactivity falls out of the graph. You don't build it.
Built on DuckLake
Every node output (SQL results, Python outputs, materialized tables) lands in DuckLake as a Parquet table on cache-fronted object storage. Fast scans, snapshot isolation, no warehouse to provision, no cluster to size.
An agent that edits the graph
Not a chat sidebar. The agent reads lineage, edits the DSL, runs nodes, and sees what broke. Every change is an ordinary diff you can review, undo, or take over.
Bring your editor
Panels speaks MCP, so Claude Code, Codex, Cursor, or anything else that speaks the protocol can drive your project directly. Same primitives, same graph, no lock-in to a single model or surface.
Start free, scale when you need to
Free
Everything you need to start
- 3 projects
- Unlimited seats
- 100 compute minutes / mo
- 5 GB storage
- 25 AI credits / mo
Pro
For production workloads
Everything in Free, plus:
- Unlimited projects
- Unlimited seats
- 1,000 compute minutes / mo
- 50 GB storage
- 300 AI credits / mo
- Pay-as-you-go overage
Enterprise
For teams that need more
Everything in Pro, plus:
- Custom compute & storage
- Custom AI credits
- SSO & audit logs
- Dedicated support
No credit card required.
Need SSO, custom limits, or something else? Get in touch