nao is an open source framework that lets data teams build and deploy AI-powered analytics agents. It connects directly to your data warehouse, ingests your full schema automatically, and reads your dbt project manifest to understand model lineage. You create your agent's context using the nao-core CLI, adding data, metadata, docs, tools, and MCPs, then deploy a chat interface so anyone on your team can run analytics on company data. nao is self-hosted, so your data stays in your infrastructure.

• Open Context Builder: Create a file-system-like context for your analytics agent. Add data, metadata, docs, tools, and MCPs with no limits. Version and test your context before deploying.
• Native Data Warehouse Integration: Connects directly to BigQuery, Snowflake, Databricks, PostgreSQL, and DuckDB/MotherDuck. Automatically ingests tables, columns, types, relationships, and cardinality.
• Agent Reliability Visibility: Unit test your analytics agent before deployment and track its performance over time so you can validate accuracy and catch issues early.
• Self-Hosted and Secure: Run everything on your own infrastructure using your own LLM API keys (OpenAI, Anthropic, Bedrock, or Gemini). No data is sent to an LLM without explicit permission.
• Built-In Feedback Loop: Users can mark answers as correct or incorrect. That feedback feeds back into context ranking to continuously improve agent accuracy.
• Natural Language to Insights: You can ask questions in plain English to receive direct analytical answers. This eliminates the need for users to write complex SQL or code.
• Native Data Visualization: The platform allows you to generate and customize visualizations within the chat interface, enabling you to view results through dynamic charts and tables.
• Transparent Reasoning: When the agent answers a question, you can review its logical process. This feature shows exactly how the agent arrived at its conclusion and identifies the specific sources it used.

• Purpose-Built for Data Work: Unlike general-purpose AI code editors, nao understands data schemas, lineage, and warehouse metadata natively, so suggestions are grounded in your actual data.
• Full Transparency: Every query shows the generated SQL, which tables were queried, when data was last refreshed, and a confidence level. No black box.
• Strong Security Model: Self-hosted with your own LLM keys means sensitive data never leaves your infrastructure without explicit permission.
• Data Stack Agnostic: Works with any warehouse, any LLM, and any context type, so you don't have to rethink your existing stack.

• Initial Learning Curve: Advanced features like context engineering, agent testing, and MCP integrations take time to set up and understand fully.
• Narrow Target Audience: Built specifically for data professionals (analysts, engineers, data scientists). Not useful for general-purpose coding or non-data workflows.

• Ad-Hoc Business Analytics: Business users can ask questions about company data in natural language without waiting for the data team to build a dashboard. The agent generates SQL, runs it against your warehouse, and returns results with full transparency into the query logic.
• dbt Model Development: Data engineers can use nao to generate, preview, and test dbt models directly from their warehouse context. The AI understands your dbt manifest, lineage, and column definitions, so you spend less time on boilerplate and more on modeling decisions.
• Data Quality Monitoring: Run data diffs between development and production, execute predefined tests, and catch breaking changes before they reach production. The agent flags issues and suggests fixes based on your schema context.
• Onboarding New Data Team Members: New hires can explore the data warehouse through the chat interface, ask questions about table relationships and business logic, and get grounded answers based on actual schema and documentation rather than outdated wiki pages.
• Replacing Ad-Hoc BI Requests: Instead of fielding one-off Slack requests for data pulls, deploy nao alongside your existing BI stack. Dashboards handle KPI monitoring while the agent handles exploratory analysis and custom queries.
• Agent Reliability Engineering: Data teams can unit test their analytics agent against a labeled set of questions, track accuracy and latency metrics over time, and version their context files with git to maintain a production-grade analytics experience.