Conversational analytics on the Databricks Intelligence Platform. Genie spaces are scoped to curated datasets, learn from feedback and example queries, and expose usage monitoring for data teams.

Generative SQL in the Amazon Redshift query editor and natural-language analytics in Amazon QuickSight (Q), producing queries and visuals from plain-English questions.

Copilot experiences across Fabric workloads — natural language to SQL in the SQL analytics endpoint and warehouse, DAX generation in Power BI, and notebook assistance in Data Engineering.

Managed text-to-SQL service inside Snowflake Cortex. Business users ask questions in natural language; a semantic model (YAML) grounds generation against governed definitions before SQL touches the warehouse.

Gemini-powered assistance across BigQuery: natural-language query generation in BigQuery Studio, data canvas for visual NL-driven exploration, SQL completion and explanation.

MIT DuckDB/ANSI text-to-SQL engine that maps governed metrics to deterministic SQL, discovers bridge tables with a NetworkX Steiner tree, and validates ASTs with SQLGlot.

Agentic text-to-SQL system that plans, selects schema, executes against a physically read-only database, repairs errors, verifies semantics and exposes traces plus an MCP server.

MIT-licensed reference BI stack joining configurable multi-source ingestion, a DuckDB star schema, dashboards, workflow rules, and a Claude text-to-SQL/RAG assistant. It excludes PII from the reporting layer, redacts questions, parses model SQL with sqlglot, restricts execution to approved read-only tables, caps rows, and records tool calls in an append-only audit log.

Beginner-oriented PostgreSQL text-to-SQL notebook that returns tables or basic charts, maintains conversation context, retries SQL errors, and layers keyword validation over read-only transactions and a statement timeout.

Open-source text-to-SQL analytics agent with governed metrics, abstention, SQL guardrails, PII pseudonymization, BIRD Mini-Dev evaluation, red-team artifacts, FastAPI and MCP serving.

Open-source text-to-SQL analytics agent with a governed metric layer, confidence-based abstention, read-only SQL guardrails, PII pseudonymization, FastAPI/MCP serving, and reproducible BIRD Mini-Dev plus Olist evaluations.

Self-hosted text-to-SQL assistant for PostgreSQL, MySQL and SQLite that returns answers with the executed SQL, using FastAPI, React, local embeddings and sqlglot-based guardrails.

Governed NL2SQL agent built around AST rewriting, database privileges, cost caps, semantic metrics, audit replay and a deliberately bounded execution pipeline.

MIT-licensed governed analytics reference stack using dbt MetricFlow, a constrained MCP metrics interface and a benchmark over public US federal procurement data.

Auditable text-to-SQL demo over synthetic portfolio datasets. It supports model-generated SQL or a deterministic, keyless compiler, runs both through structural verification and a read-only guard, and shows the SQL beside every result.

Company-facts and semantic-context system for AI agents that separates live surveyed data, human-approved facts and source records, exposes answers through MCP, and can run self-hosted in a VPC.

Self-hostable AI data analyst for SQL warehouses and REST APIs, grounded in an authored YAML semantic layer with review-gated knowledge context, MCP, dashboards and embeddable clients.

Grounded analytics and text-to-SQL platform built around governed metrics, lineage and change management, with database adapters and OAuth-protected MCP access.

Apache-2.0-style reference lakehouse with an integrated local text-to-SQL layer over DuckDB and Parquet, schema/metric retrieval, parser-based table allow-listing, forced row limits, a locked external-file sandbox and one-step SQL repair.

Open-source kit for assembling reviewable analytics-agent context from dbt models, warehouse schemas, dashboards, query history and documentation while preserving evidence and surfacing unresolved meaning as questions.

Photovoltaic-domain conversational BI system with retrieval, text-to-SQL, layered validation, read-only execution, self-correction and committed development and held-out evaluations.

MIT agentic SQLite text-to-SQL system with LangGraph correction loops, SQLGlot validation, local RAG, CLI/Streamlit interfaces and committed BIRD Mini-Dev ablations.

MIT-licensed process-industry natural-language analytics curriculum and reference stack using PostgreSQL semantic views, SQLGlot guardrails, read-only roles, row-level security, golden-answer generation and evaluation tooling.

Self-learning data agent that combines six context layers, validated-query knowledge, error learnings, live schema introspection, and separate analyst/engineer roles for answering business questions over PostgreSQL.

Small LangGraph/LangChain text-to-SQL agent for SQLite. It introspects tables, columns and foreign keys, generates one read-only query, validates it before execution, opens SQLite in read-only mode, caps results at 100 rows and retries failures up to three times. The repository includes offline tests and a 20-question exact-result harness; its reported 20/20 agent run is on the bundled sample database and may vary by model.

Apache-licensed enterprise data-agent framework combining NL2SQL, semantic retrieval, multi-agent workflows, feature engineering, YAML orchestration, and MCP/A2A tool integration.

Open-source natural-language-to-SQL service for interacting with SQL databases using LLMs; the public repository is mature but has seen no code push since July 2024.

Agentic text-to-SQL analytics application built with LangGraph, FastAPI and React for PostgreSQL and CSV data, with SQLGlot-based validation, statistical-outlier detection and streamed traces.

Chinese-language NL2SQL application for job-market data with three-stage validation, self-repair, visualization, and a committed 102-question exact-result evaluation.

Self-hosted conversational BI platform for natural-language analysis across CSV, Parquet, Excel and PostgreSQL, with DuckDB execution, charts, dataset-level access control, scheduled reports, Skills and MCP extensions.

Spanish-language governed NL-to-SQL lakehouse agent prototype. Its implemented layers parse SQL into an allowlisted AST, enforce role/column/position policies, and estimate query cost from Iceberg manifests before DuckDB execution; generation, masking, and audit layers are not yet complete.

Local-first Windows desktop database assistant that turns natural-language requests into typed, reviewable operations, with physically read-only query paths and explicit one-use confirmation for SQLite writes.

MIT-licensed agentic data assistant and application platform that connects to databases and files, generates and executes SQL and Python analysis, runs reusable skills, and produces reports and dashboards.

Local-first Windows database agent for natural-language SQL, charts, and controlled writes across SQLite, MySQL, and PostgreSQL, with read-only defaults, typed reviewable operations, bounded execution, and one-use write confirmation.

MIT-licensed database client with 3D and 2D ER diagrams, SQLite/PostgreSQL support, editable data views and text-to-SQL generation through OpenAI-compatible local or cloud model endpoints.

Self-hosted Postgres and MySQL database agent for natural-language BI, schema exploration, slow-query analysis, index recommendations, dashboards and MCP access, with bring-your-own-model endpoints.

Agent-native analytics-engineering toolkit for warehouse exploration, dbt transformation and semantic modeling, schema-drift maintenance, and guarded verification of generated changes.

Governed serving layer between AI agents and data warehouses, returning definition-grounded answers with SQL, confidence grades and signed receipts through MCP or REST.

MIT-licensed, Olist-specific analytics stack combining DuckDB views, a metric semantic layer, dashboards and a deterministic natural-language router that selects allowlisted SQL templates rather than generating arbitrary SQL.

Apache-2.0 knowledge-augmented enterprise agent prototype combining intent routing, RAG and read-only DuckDB text-to-SQL. The maintainer explicitly marks knowledge-graph, claim-evidence verification, evaluation and frontend work as unfinished.

Cyclic LangGraph text-to-SQL reference with SQLGlot validation, independent read-only SQLite/PostgreSQL enforcement, timeouts, row caps and a bounded typed-error repair loop.

Early governed text-to-SQL analytics engine for ClickHouse with domain packs, tracing and a committed layer-level evaluation; still missing a declared software license and several planned controls.

Evidence-first game-operations incident investigation agent that exposes metric definitions, read-only SQL, cohort comparisons, anomaly detection and citation-checked reports through MCP tools.

Self-hosted generative BI in a single Go binary, with Excel import, text-to-SQL, dashboards, RBAC, data subjects, branded portal and local SQLite storage.

Self-hosted generative BI platform that imports Excel into isolated SQLite-backed data subjects and lets an LLM build and revise dashboards. Ships as one Go binary with embedded React UI, read-only SQL execution, page versioning, ACLs, RBAC and render-feedback checks.

Role-aware HR analytics assistant for MySQL/SQLite with schema grounding, SQLGlot validation, read-only execution, clarification, repair, follow-ups and charts.

Self-hosted HR analytics assistant for MySQL/SQLite with role-scoped schema access, sqlglot query validation, read-only database credentials, bounded repair, charts, and a 55-question execution-accuracy harness.

MCP SQL privacy gateway that parses agent-written queries, traces outputs to protected base columns, regenerates read-only SQL and replaces declared personal values with stable local tokens. A signed-in human can resolve tokens locally; the model cannot.

MIT-licensed, self-hosted business analytics agent that turns constrained natural-language plans into validated read-only SQL, auditable reports and MCP/FastAPI interfaces. The repository includes a 20-case demo regression set, retrieval metrics, tests and explicit limits on what has not been validated with real models or services.

Chinese self-service analytics agent with a governed semantic layer, BM25 retrieval, SQL AST allowlisting, read-only execution, trace replay and reproducible result-equivalence evaluation. Its published ablations explicitly report that the critic did not improve SQL execution accuracy.

MIT Laravel NL2SQL package and Blade chat widget for PostgreSQL, MySQL, MariaDB and SQLite. Sends schema structure rather than rows to the configured model, supports hosted or local models, and exposes benchmark, privacy and validator caveats.

Deterministic schema-context map for data analysts and text-to-SQL agents, serving join paths, keys, grain and status codes through MCP, CLI JSON or Markdown without LLM calls.

MIT-licensed browser database studio supporting many SQL, NoSQL and infrastructure engines, with SSO, audit trails and AI-assisted query generation.

MIT-licensed synthetic seller-operations agent demo with fixed-intent text-to-SQL over five SQLite read-only views, an authorizer-backed allowlist, row caps, audit chaining, PII-scrubbed feedback and human confirmation gates. Its SQL component is template-routed rather than free-form generation.

Self-hosted, metric-first conversational analytics agent that constrains SQL generation with published metric SQL, themes, models and join relationships; asks for clarification when semantic retrieval misses.

Apache-2.0 LangGraph data agent that routes natural-language questions across Text-to-SQL, PostgreSQL/PGVector RAG and Neo4j graph queries, then unifies results through an evidence-lineage contract and observable execution trace.

Open-source analytics-agent framework with a CLI for building versioned context, a self-hosted chat interface, evaluation commands and feedback workflows for deploying natural-language analytics to teams.

MIT-licensed schema-grounded SQLite text-to-SQL prototype with foreign-key join-path hints, SQLGlot read-only validation, a two-round repair loop, execution traces and a committed 30-question result-matching evaluation.

Conversational Gemini NL2SQL application for SQLite and PostgreSQL with schema retrieval, English/Tamil/Thanglish input, voice, charts, self-correction and staged write confirmation.

Deterministic semantic-layer and ontology drift auditor that checks ontology JSON against warehouse DDL, SQL templates and optional CSV samples, producing CI-friendly findings and scores without LLM calls.

Open-source conversational BI and Text2SQL agent with schema linking, catalog retrieval, persistent memory, Plotly visualization, MCP support, and Streamlit/Gradio interfaces across multiple warehouses and LLM providers.

Open-source prototype for Korean brokerage support: guarded text-to-SQL over overseas-stock trades, deterministic capital-gains calculations, local regulatory RAG, and human-review gating. SQL is AST-validated, customer-scoped, read-only, time-limited and row-capped.

Alpha Python regression-test and CI-gate toolkit for text-to-SQL and analytical agents. Runs generated SQL against SQLite fixtures and checks golden-result, invariant, or baseline-differential expectations before returning a failing exit code.

Open-source English/French pharmaceutical sales NL2SQL reference that resolves entities before SQL generation, requires users to approve interpretations, validates SQL with an AST guard, and isolates generated Python in a constrained child process.

Local-first macOS conversational-BI desktop app that delegates planning to installed CLI agents, requires them to emit validated intermediate representations rather than SQL, then compiles and executes read-only queries locally.

Adaptable MCP reference server for governed warehouse access: validates model SQL, injects tenant predicates, masks PII, estimates cost and independently re-parses the final statement before read-only execution.

Security-focused text-to-SQL reference implementation for fictional SQLite ecommerce data. It treats model output as untrusted, validates SQL with SQLGlot, injects row-level authorization server-side, and ships correctness and adversarial regression suites.

Open-source PostgreSQL text-to-SQL application with layered read-only controls, schema retrieval, self-correction, Redis caching and a published 100-question BIRD mini-dev ablation study.

Chinese-language NL2SQL agent for BI workflows, combining terminology normalization, schema linking, model routing, SQLGlot validation, read-only execution, result auditing and bounded self-correction.

MIT-licensed Chinese-language NL2SQL agent for business intelligence with business-term normalization, deterministic schema linking, SQLGlot validation, risk scoring, read-only execution, result auditing and bounded self-correction.

Ambiguity-aware PostgreSQL/SQLite text-to-SQL prototype that offers explicit business-meaning choices before generation and applies AST validation, table whitelisting and read-only execution.

Open-source PostgreSQL natural-language analytics app that generates read-only SQL, returns results and rule-selected charts, and evaluates the same production pipeline with a 123-question execution-based harness.

Graph-powered Text2SQL application and SDK that models database schemas as a graph, generates and executes SQL from natural language, and exposes REST, MCP, Python SDK, Docker, and multi-turn interfaces for PostgreSQL and MySQL.

Apache-2.0 enterprise data-agent stack combining governed metadata and knowledge graphs, reusable analysis skills, and a controllable artifact-oriented host runtime. v0.3.0 adds headless HTTP/replayable SSE execution, scheduled runs, snapshots, forks and host-owned delivery.

MIT-licensed MCP server for guarded read-only SQLite queries, with row and time limits, audit logging, object-name hints and a model-through-server text-to-SQL evaluation harness.

Read-only database-agent toolkit for SQL review, runbook Q&A and live PostgreSQL diagnosis, with deterministic guardrails around agent-generated operations.

MIT LangGraph text-to-SQL reference agent for a bundled music-store schema. It combines SELECT/WITH-only execution, retries on executor or guardrail errors, a persistent playbook of failure-derived lessons, optional Mem0 user memory, Streamlit UI and MCP execution tool.

Apache-2.0 work-in-progress semantic intelligence platform combining a business semantic layer, capability registry, governed read-only execution, MCP client/server support, and NL2SQL/ChatBI interfaces. The repository contains backend routes, migrations, query/capability modules, tests and deployment scaffolding, but the README labels the project WIP and adoption is currently negligible.

Chinese-language educational NL2SQL data agent implementing metadata ingestion, vector and full-text retrieval, LangGraph orchestration, SQL generation/validation/execution, SSE streaming, and a React interface for an e-commerce warehouse.

MIT-licensed MCP analytics server that lets coding agents produce browser or terminal dashboards from read-only PostgreSQL, SQLite, Excel, and CSV queries.

Inspectable Python text-to-SQL research agent comparing one-shot, tool-using and database-experienced Claude workflows on a cleaned BIRD subset, with an MCP database server, read-only execution, reproducible splits and tests.

Family of open-weight text-to-SQL models (SQLCoder) plus a commercial platform. SQLCoder models were repeatedly competitive with much larger general models on SQL generation benchmarks.

MIT-licensed self-correcting NL2SQL reference implementation for Chinese-language BI workflows, combining business-term normalization, schema linking, model routing, SQLGlot validation, explain-based scan checks, read-only execution, result auditing and bounded retries.

Apache-2.0 text-to-SQL workbench and Python SDK with configurable schema retrieval, few-shot prompting, ORM query objects, evaluation and budgeted grid search across SQLite, MySQL and PostgreSQL.

Deterministic semantic SQL inspector that checks AI- or human-written queries for fan-out, additivity, join-key, cross-join, weighted-average, and sensitive-column violations before execution.

Developer-preview governed NL2SQL platform that keeps context and generated SQL inspectable, validates fail-closed, requires explicit approval and runs through bounded read-only connections.

Local, open-source SQLite text-to-SQL agent that runs against an OpenAI-compatible endpoint, repairs SQL errors, generates sandboxed charts, records JSONL traces, and exposes ask/query/schema tools through MCP.

Python SDK for agentic text-to-SQL that lets a model explore schemas and test/self-correct queries through an execute_sql tool, with dialect adapters, tracing, optional scenarios, MCP trace analysis, and LangChain middleware.

MIT-licensed Chinese natural-language-to-SQL reference that routes generated SQL through a table, column and function allow-list, enforces row limits, and executes against a read-only SQLite sandbox with an authorizer and timeout.

Open-source Python text-to-SQL agent built on Google ADK. Extracts SQLite or PostgreSQL schemas, generates SQL with Gemini, validates syntax and referenced schema objects with SQLGlot, executes queries, and retries failed SQL up to three times.

A controllable enterprise-oriented text-to-SQL reference implementation built with LangGraph, FastAPI, SQLGlot, Qdrant, and Elasticsearch. It combines semantic preflight, multi-path metadata retrieval, AST-based SQL safety checks, bounded correction, recovery flows, SSE observability, and packaged evaluation suites.

MIT-licensed Claude-powered DuckDB analytics agent with schema exploration, bounded tool use, read-only execution, self-correction and a 12-question execution-accuracy evaluation harness.

Local MySQL text-to-SQL prototype using Ollama and Qwen2.5-Coder, with an ambiguity check that can ask follow-up questions before generating SQL, a Streamlit UI, schema viewer, and result summaries.

MIT-licensed Python text-to-SQL research implementation that maps questions into a semantic query, resolves business concepts and join paths through Neo4j, deterministically compiles SQL, validates and guardrails it, then executes against PostgreSQL.

Experimental Python text-to-SQL interface focused on AST validation, schema-reference checks and read-only execution controls.

Early Python text-to-SQL prototype with schema retrieval, multi-model BYOK support and a self-correction loop.

MIT-licensed, user-aware data-agent framework for natural-language-to-SQL, database tools, streaming tables/charts and permission-aware execution. The main GitHub repository is archived; its docs and PyPI package remain online, with v2.0.2 as the latest release.

Previously published end-to-end Snowflake Cortex Agents reference stack. Its GitHub repository has returned 404 on repeated checks since at least September 12, 2026, so the implementation is currently unavailable for verification.

Agent-oriented generative BI and governed text-to-SQL system built around an MDL semantic layer, versioned business context, execution guardrails, and connectors for more than 20 data sources.

Governed generative-BI and text-to-SQL system built around an open context layer, producing SQL, charts, dashboards, and agent-facing context across 20+ data sources.

Local and self-hosted governed semantic layer for AI database analysis, with human-approved metrics and joins, snapshot-pinned YAML models and receipts recording SQL, relationships and model version.

Cross-platform, local-first database client and SQL workspace supporting 40+ databases, bring-your-own-model SQL generation/explanation/optimization, desktop/web/Docker/CLI delivery, and MCP.

Source-available, single-user self-hosted analytics assistant combining natural-language SQL, governed context, dashboards, metrics, monitoring and execution-time scope enforcement.

Source-available semantic sidecar that compiles version-controlled YAML metrics and dimensions into SQL across eight dialects, exposing governed queries over REST, PostgreSQL wire, Flight SQL and MCP. v2.27.1 adds a tested DuckDB ADBC client path and fixes Flight SQL schema negotiation.

Source-available self-hosted BI application with natural-language querying, AI-assisted SQL, semantic metrics, dashboards, row-level security and MCP support in a single Go binary.

Source-available, self-hosted BI platform in Go and React with natural-language analytics, SQL assistance, semantic metrics, dashboards, row-level security, audit logs and five read-only MCP tools. The repository documents SQL guards, result masking and identity-bound MCP calls.

Apache-2.0 evaluation harness for the table-retrieval stage before NL2SQL. It compares six BigQuery and Dataplex discovery strategies across a 15-table corpus, 25 graded questions and five repetitions, with Vertex AI Pipelines, checkpointing, scoring and committed full-run results.

Runnable ClickHouse text-to-SQL agent using Claude tool calls, structured answers, a database-level read-only user, an application single-SELECT guard, semantic notes, and deterministic evaluation against SQL-computed answers.

Chinese-language photovoltaic analytics prototype that routes table and metric context into text-to-SQL, validates and executes queries through a read-only SQLite connection, retries bounded execution failures, and streams traces over SSE. Includes a 31-question synthetic-domain evaluation and committed per-version outputs. No software license is declared.

Experimental multi-agent analytics stack combining hybrid schema retrieval, SQL generation, sqlglot AST policy checks, DuckDB execution, statistical diagnostics and numeric groundedness auditing.

Experimental Python text-to-SQL analyst for cross-border e-commerce with a committed 84-question Chinese evaluation set and eight ablations; the repository reports business-context grounding improving its own evaluation from 33.3% to 94.0%.

A Chinese-language research implementation for text-to-SQL over a seven-table SQLite retail warehouse. It includes read-only validation, execution timeout, optional repair and self-check loops, schema retrieval, and an 84-question executable evaluation set. The repository reports that adding business metric definitions to schema context raised result accuracy from 33.3% to 94.0%; it has no declared software license.

Domain analytics agent for the Olist e-commerce dataset with metric definitions, SQLGlot safety controls, deterministic calculations, FastAPI/Streamlit interfaces and a frozen 60-case evaluation.

Production-oriented natural-language analytics system over Moraware and QuickBooks data. Routes questions across read-only SQL, semantic retrieval, hybrid SQL-scoped retrieval, and refusal paths; publishes a sanitized 82-question evaluation and operational measurements.

Reproducible Phi-3 text-to-SQL QLoRA training and evaluation workbench with paired base/adapter predictions, bounded SQLite execution scoring, published manifests, a recorded evidence viewer, and optional model serving.

ReAct text-to-SQL research implementation for SQLite that verifies numeric answer claims against returned rows and ships a BIRD mini-dev evaluation harness with intervention and recovery metrics.

Previously tracked early SQL guardrail and MCP reference implementation. The recorded GitHub repository was unavailable at verification time.

Research artifact for calibrated pre-execution data-minimization guards for LLM database agents, with deterministic SQL facts, role policies, conformal risk control, workloads, scores and human labels.

Databricks-focused text-to-SQL evaluation harness that tests answerable, ambiguous and unanswerable questions, including whether a system should refuse unsupported requests.

Reproducible MIT evaluation of QLoRA fine-tuning for Qwen2.5-Coder-3B on BIRD Mini-Dev, measuring capability (pass@10), repeatability (pass^10), failure modes and comparison with a prompted 7B model.

Research implementation for multi-step NL2SQL and data analysis with clarification, schema retrieval, read-only SQL controls, sandboxed Python computation, provenance traces, and reproducible evaluation scripts.

Experimental semantic-first conversational BI client exploring model-agnostic LLM configuration, database schema discovery and a Cube-based semantic layer; natural-language querying remains roadmap work.

Uber's internal natural-language-to-SQL system, described publicly in engineering posts: multi-agent pipeline with workspace routing, intent classification, and iterative self-correction over Presto-scale schemas.

PostgreSQL chat-to-SQL prototype that separates model generation from deterministic SQLGlot validation and layered read-only execution, with allow-listed tables, limits, timeouts, audit rows, a web UI and an MCP server.

Configuration benchmark comparing raw SQL, glossary-assisted SQL, dbt MetricFlow and Malloy interfaces with and without an agent harness over one frozen analytics warehouse.

MIT-licensed text-to-SQL agent evaluation rig with a 192-task benchmark, audited grader, paired McNemar analysis and per-score trace replay.

Offline benchmark for text-to-SQL verification. It executes gold and faulty SQL against a local engine, compares verification detectors, demonstrates data-scope sensitivity, and includes a destructive-operation guardrail comparison.

Full-stack supply-chain analytics platform with a tool-calling copilot, RAG and an allowlisted read-only text-to-SQL layer over MySQL. The repository documents synthetic-demo provenance, backend tests and explicit deployment and RBAC limitations.