Skip to content

Chat with Your Data

An open-source Python framework for building extensible, domain-aware AI data applications.

Lumen is designed for teams who want conversational data exploration without giving up control:

  • Open source and inspectable end-to-end
  • Extensible and configurable: plug in custom data sources, Python analyses, tools, and agents
  • Domain-aware: encode business logic, terminology, and constraints directly in code

Ask questions in natural language, but keep execution explicit, reproducible, and grounded in Python.

Quick Demo

New to Lumen? Try our 30-second quick start → to see it in action with a walkthrough.

Bring your own LLM. Then connect to your data sources: Snowflake, BigQuery, DuckDB, and any database supported by SQLAlchemy. See all installation options →

export OPENAI_API_KEY=sk-...

lumen-ai serve your_data.csv --provider openai --show  # parquet also works

Then ask:

  • Show me the top 10 rows
  • What's the average revenue by region? Plot as a bar chart.
  • Filter for sales over $1000 and create a histogram
from lumen.sources.duckdb import DuckDBSource
import lumen.ai as lmai

source = DuckDBSource(
    tables={
        "penguins": "https://datasets.holoviz.org/penguins/v1/penguins.csv"
    }
)

llm = lmai.llm.Bedrock()  # LiteLLM also works
ui = lmai.ExplorerUI(data=source, llm=llm)
ui.servable()

Then ask:

  • Which islands have the most penguins? Plot as a horizontal bar chart.
  • Create a scatter plot of bill length vs body mass, colored by species
  • Calculate the ratio of flipper length to body mass, then plot it
from lumen.sources.sqlalchemy import SQLAlchemySource

source = SQLAlchemySource(
    url='sqlite:///data.db'  # any database flavor
)
llm = lmai.llm.AINavigator()  # LlamaCpp or Ollama also work
ui = lmai.ExplorerUI(data=source, llm=llm)
ui.servable()

Then ask:

  • How much sleep did I get last week?
  • Show a timeseries of my daily step counts
  • What's the average heart rate during workouts?
from lumen.sources.bigquery import BigQuerySource
import lumen.ai as lmai

source = BigQuerySource(project_id='your-project-id')

llm = lmai.llm.Google()
ui = lmai.ExplorerUI(data=source, llm=llm)
ui.servable()

Then ask:

  • What are the peak traffic hours? Show as a line chart.
  • Calculate conversion rate by source and display in a table
  • Find anomalies in daily user counts
from lumen.sources.snowflake import SnowflakeSource
import lumen.ai as lmai

source = SnowflakeSource(
    account='your-account',
    database='your-database',
    authenticator='externalbrowser'
)

llm = lmai.llm.MistralAI()
ui = lmai.ExplorerUI(data=source, llm=llm)
ui.servable()

Then ask:

  • Which customers have the highest lifetime value?
  • Show monthly revenue trends for the last year
  • Join orders and products, then show top categories by profit
export OPENAI_BASE_URL=http://localhost:11434/v1
export OPENAI_API_KEY=ollama

lumen-ai serve postgresql://user:pass@localhost/mydb \
  --provider ollama \
  --model qwen3:32b \
  --show

Then ask:

  • Show me customer signup trends by month
  • Which products have declining sales? Show the trend.
  • Find users who haven't logged in for 90 days

Why Lumen?

Stop waiting for data teams. Ask questions in plain English.

You have questions about your data. Your data team is backlogged three sprints. You could learn SQL, but by the time you do, the question will have changed.

Lumen eliminates this bottleneck. See launch options →

Who Benefit
Data analysts Skip boilerplate code and focus on insights
Developers Build data apps in hours instead of weeks
Data teams Enable self-service without losing control
Researchers Share reproducible analysis without teaching syntax

No vendor lock-in. Use OpenAI, Google, Anthropic, Mistral, or run models locally with LlamaCPP so that your data never leaves your machine. Configure your LLM provider →

# Run 100% local - no API keys, no cloud, full privacy
llm = lmai.llm.LlamaCpp(
    model_kwargs={
        'default': {
            'repo_id': 'unsloth/Qwen3-8B-GGUF',
            'filename': 'Qwen3-8B-Q5_K_M.gguf'
        }
    }
)
ui = lmai.ExplorerUI(data='data.csv', llm=llm)

# Or switch to cloud providers anytime - just one line
# llm=lmai.llm.OpenAI() or Anthropic(), Google(), MistralAI()

Connect to anything. PostgreSQL, Snowflake, DuckDB, or CSV files—if it has data, Lumen can support it. See all data sources →

# Mix sources - local files + cloud databases
ui = lmai.ExplorerUI(data=[
    'local_data.csv',                                    # CSV file
    'postgresql://localhost/mydb',                       # PostgreSQL
    SnowflakeSource(account='...', database='...'),      # Snowflake
    BigQuerySource(project_id='my-project')              # BigQuery
])

Designed for extension. Pure Python + YAML configs. Build custom agents, tools, and visualizations when you need them. Or use Reports to turn recurring analyses into one-click reproducible dashboards. View tutorials →

# Add domain-specific analysis in minutes
class RevenueAnalysis(lmai.analysis.Analysis):
    """Calculate key revenue metrics."""

    columns = ['revenue', 'churned']

    def __call__(self, pipeline, ctx):
        df = pipeline.data
        return pd.DataFrame({
            'MRR': df['revenue'].sum() / 12,
            'Churn Rate': (df['churned'].sum() / len(df)) * 100
        }, index=[0])

ui = lmai.ExplorerUI(data=source, analyses=[RevenueAnalysis])

Analyses run deterministic calculations with code instead of LLM generation—perfect for financial metrics, scientific formulas, or any calculation that must produce identical results every time.

Learn about analyses →

# Teach agents your domain terminology
context = """
{{ super() }}

In our business:
- "Accounts" means customer accounts
- Q1 = Jan-Mar, Q2 = Apr-Jun, Q3 = Jul-Sep, Q4 = Oct-Dec
- "Active" means logged in within 30 days
"""

agent = lmai.agents.ChatAgent(
    template_overrides={"main": {"context": context}}
)
ui = lmai.ExplorerUI(data=source, agents=[agent])

Agents are specialized workers that handle different question types (SQLAgent writes queries, VegaLiteAgent creates charts). Customize their prompts to teach them your domain terminology and business rules.

Learn prompt customization →

# Extend agents with custom functions
def calculate_customer_ltv(customer_id: str, table) -> dict:
    """Calculate customer lifetime value."""
    customer_data = table[table['customer_id'] == customer_id]
    total_revenue = customer_data['revenue'].sum()
    return {
        'ltv': total_revenue,
        'avg_order': total_revenue / len(customer_data)
    }

tool = lmai.tools.FunctionTool(
    function=calculate_customer_ltv,
    requires=['table'],
    provides=['ltv', 'avg_order']
)
ui = lmai.ExplorerUI(data=source, tools=[tool])

Tools let agents access external data and perform specialized tasks—wrap APIs, integrate libraries, or add custom business logic that agents can call when needed.

Learn about tools →

What Makes It Different?

AI + YAML + Python. Use what fits your workflow.

Traditional BI Code-First Lumen
Inflexible, expensive Too much boilerplate Chat for instant results
Vendor lock-in Manual UI building YAML specs for reproducibility
No customization Full control Python extensions when needed

Chat with your data to generate visualizations instantly, write YAML specs for reproducible dashboards, or drop into Python for custom logic:

You: "Create a dashboard showing penguin measurements by species with filters"

Lumen AI generates:

  • SQL queries to aggregate data
  • Interactive filters for species, island, sex
  • Multiple visualizations (scatter plot, histogram, table)
  • Responsive layout

Try it now →

sources:
  penguins:
    type: file
    tables:
      data: https://datasets.holoviz.org/penguins/v1/penguins.csv

pipelines:
  filtered:
    source: penguins
    table: data
    filters:
      - type: widget
        field: species
      - type: widget
        field: island

layouts:
  - title: Palmer Penguins Dashboard
    pipeline: filtered
    views:
      - type: hvplot
        kind: scatter
        x: bill_length_mm
        y: bill_depth_mm
        color: species
      - type: hvplot
        kind: hist
        y: body_mass_g
      - type: table

Learn YAML specs →

Palmer Penguins Dashboard

Interactive dashboard with:

  • 📊 Scatter plot showing bill measurements by species
  • 📈 Histogram of body mass distribution
  • 📋 Filterable data table
  • 🎛️ Sidebar widgets for filtering
  • 🎨 Dark theme
  • 📱 Responsive design

Built in 15 minutes with YAML or 2 minutes with AI chat.

Community & Support

Questions? Join our community: