§ 01
Vendor-neutral
No sponsored placements. No paid rankings. Vendors cannot pay to alter catalog entries or benchmark scores.
§ Charter
Open benchmarks, a vendor-neutral tool catalog, and a common vocabulary for teams evaluating systems that query data through an LLM. Practitioner-led and maintained in the open.
§ 01 · Charter
These commitments govern what the project publishes and what it refuses to publish.
No sponsored placements. No paid rankings. Vendors cannot pay to alter catalog entries or benchmark scores.
Every benchmark ships with its prompts, traces, evaluation criteria, and a reproducible run script. Results are dated.
Maintained by working analytics engineers and data practitioners. Contributions happen through pull requests, not press cycles.
The reference architecture privileges open specifications: MCP, MetricFlow, Iceberg, OpenLineage. Proprietary tooling is documented, not recommended by default.
§ 02 · The stack
Agentic analytics composes five layers with clear responsibilities. Intent flows down from the user to the warehouse; results and reasoning flow back up. Pick a layer to go deeper.
The substrate everything else depends on: warehouses, lakehouses, and open table formats. Format openness determines tool coverage; closed catalogs and non-standard SQL dialects compound portability debt invisibly until the engine migration exposes it.
The contract between business meaning and database schema. Metric definitions, dimensional relationships, and machine-readable APIs the agent queries before writing SQL. The single highest-leverage dependency: without it, valid SQL produces semantically wrong answers.
Access control, data quality validation, lineage, and audit over the data and semantic layers. Human analysts bring implicit governance to every query; agents bring none. This layer encodes the rules an agent must operate within, so errors do not amplify at machine speed.
The agent runtime. It plans the investigation, dispatches tool calls against the layers below, evaluates intermediate results, and decides whether to continue, branch, or conclude. This feedback loop is what distinguishes agentic analytics from single-shot text-to-SQL.
Where agent output surfaces to humans and downstream systems: charts, alerts, notebook artifacts, dispatched actions. Agentic systems also render the plan, the queries run, and confidence signals, so the reasoning is inspectable rather than opaque.
§ 03 · Sections
Six entry points. Learn is the main reading path; the others are reference, calendar, evaluation, writing, and self-assessment.
Introduction to the agentic analytics stack. A hub for each layer, with open standards, failure modes, and tools.
Open →Indexed reference of tools across the stack. Filter by layer and license.
Open →Conferences where agentic analytics work is presented, debated, or benchmarked.
Open →Open evaluations across text-to-SQL, semantic queries, and agent orchestration. Reproducible.
Open →Long-form analysis of architecture patterns, observability, and operating agents in production.
Open →Self-assessment of your organization's agentic analytics maturity across five dimensions.
Open →§ 04 · Repositories
Curated repositories focused on analytics, BI, and the layers of the agentic analytics stack. General-purpose agent frameworks are excluded here; they're covered in the orchestration layer page.
DuckDB is an analytical in-process SQL database management system
Apache Iceberg
An open-source storage framework that enables building a Lakehouse architecture with compute engines including Spark, PrestoDB, Flink, Trino, and Hive and APIs
📊 Cube Core is open-source semantic layer for AI, BI and embedded analytics
Malloy is a modern open source language for describing data relationships and transformations.
MetricFlow allows you to define, build, and maintain metrics in code.
OpenMetadata is a unified metadata platform for data discovery, data observability, and data governance powered by a central metadata repository, in-depth column level lineage, and seamless team collaboration.
The Context Platform for your Data and AI Stack
Open, Multi-modal Catalog for Data & AI
🤖 Chat with your SQL database 📊. Accurate Text-to-SQL Generation via LLMs using Agentic Retrieval 🔄.
open-source agentic AI data assistant for the next generation of AI + Data products.
Open-source text-to-SQL and text-to-chart GenBI agent with a semantic layer. Ask your database questions in natural language — get accurate SQL, charts, and BI insights. Supports 12+ data sources (PostgreSQL, BigQuery, Snowflake, etc.) and any LLM (OpenAI, Claude, Gemini, Ollama).
Apache Superset is a Data Visualization and Data Exploration Platform
The easy-to-use open source Business Intelligence and Embedded Analytics tool that lets everyone work with data :bar_chart:
Business intelligence as code: build fast, interactive data visualizations in SQL and markdown
Text-to-SQL via RAG.
AI agent that lets data teams chat with data.
SoTA LLM for converting natural language to SQL.
Natural language to SQL engine with warehouse-grounded evaluation.
AI native data app framework.
Semantic layer for analytics APIs.
Chat-based SQL client for the next decade.
In-process analytics SQL engine.
§ 05 · Contribute