MIT open source · runs locally · bring your own key
Ask your database for a dashboard.
Describe what you want to see in a sentence. Lumyr looks at your real tables, checks its queries against your data, and builds a working Streamlit dashboard you can keep refining in plain English. It runs on your computer with your own Anthropic key, so your data goes straight from your warehouse to your screen.
E-commerce overview, 2025
Completed and shipped orders · refreshed every 10 minutes
Monthly revenue and gross profit
Top categories by revenue
A sentence in, a dashboard out
-
Connect your data
Point Lumyr at BigQuery, PostgreSQL, Snowflake, Redshift, NetSuite, anything with an ODBC driver, or drop in a spreadsheet. Credentials are encrypted with your operating system's key store.
-
Say what you want to see
Weekly signups by channel for the last quarter, with this week highlighted.
No SQL, no chart builder, no picking fields from a menu. -
Get a dashboard built on your real columns
Before writing anything, the agent reads your schema and runs read-only queries to confirm every table, column and number it plans to use. Then it builds the dashboard and Lumyr opens it.
-
Keep talking to it
Split that by region.
Make it a bar chart.
Each request edits the dashboard you're looking at, so it gets better instead of starting over.
One connection leaves your machine
There's no Lumyr account and no Lumyr server. Everything runs on your computer, and you can see exactly what goes out.
Your request, your table and column names, and any rows the agent samples while checking its queries. That's the complete list.
The output is ordinary code
Every dashboard is a readable Streamlit file. Commit it, deploy it, or edit it by hand. It keeps working after you uninstall Lumyr.
- BigQuery
- PostgreSQL
- Snowflake
- Redshift
- NetSuite
- ODBC
- CSV / Excel
The agent is open source too. Lumyr builds dashboards with opencode, running inside the app with sharing and auto-update turned off.
import streamlit as stimport plotly.express as px@st.cache_data(ttl=600)def monthly_revenue(): return query(""" SELECT DATE_TRUNC('month', created_at) AS month, SUM(sale_price) AS revenue, SUM(sale_price - cost) AS gross_profit FROM order_items WHERE status IN ('Complete', 'Shipped') AND created_at >= '2025-01-01' GROUP BY 1 ORDER BY 1 """)df = monthly_revenue()st.title("E-commerce overview, 2025")a, b = st.columns(2)a.metric("Revenue", money(df.revenue.sum()))b.metric("Gross profit", money(df.gross_profit.sum()))st.plotly_chart( px.line(df, x="month", y=["revenue", "gross_profit"]), width="stretch",)Build it from source
You'll need Node 20 or newer and an Anthropic API key. Python is downloaded for you by the setup step.
Open Settings on first launch and paste your key in. Create one at console.anthropic.com.
On a Mac, clone outside iCloud-synced folders like Documents or Desktop. iCloud can offload dependency files, which makes builds and dashboards slow to start.
# get the code
$ git clone https://github.com/gautamgupta0196/Lumyr
$ cd Lumyr && npm install
# self-contained Python for running dashboards
$ npm run prepare:python
# start the app
$ npm run devBefore you try it
What does it cost?
Lumyr is free and MIT licensed. You pay Anthropic directly for the tokens you use, at their standard rates. A full multi-chart dashboard cost about $0.30 on Claude Opus 5 in our testing; smaller requests and cheaper models cost less.
Can it change my data?
The agent is told to run read-only queries, but Lumyr doesn't enforce that. Connect it with a read-only database role. A permission your database enforces is a stronger guarantee than an instruction to a model.
Which model does it use?
Claude Opus 5 by default. Switch to Sonnet 5 or Haiku 4.5 in Settings if you want answers faster or cheaper.
Does it collect anything about me?
No. There's no sign-in and no usage tracking. Crash reporting is off unless you build the app with your own Sentry account.