Defog
Defog is a natural language data query platform that lets enterprises ask free-form questions of structured databases via its fine-tuned open-source SQLCoder LLM family. It serves enterprise customers in finance, healthcare, manufacturing, and government with privacy-first, on-premise-capable deployments.
- Company typePrivate
- Founded2023
- HeadquartersSan Francisco, United States
- Headcount1–10
- GTM typeB2B
- OfferingSoftware
What Defog does
Defog (legally Full Stack Data Pte Ltd, incorporated in Singapore with a US-facing presence at defog.ai) builds a natural language data query platform that lets non-technical users ask free-form questions of structured databases and receive generated SQL, executed queries, and visualizations in response. The platform is anchored by SQLCoder, a family of fine-tuned open-source large language models ranging from 7B to 70B parameters and built on third-party base architectures including StarCoder, CodeLlama, and Mistral. SQLCoder-70B is reported to achieve 93% accuracy on text-to-SQL tasks for schemas not seen during training, with reported accuracy exceeding 99% after per-customer fine-tuning. The product surface extends beyond single-query text-to-SQL into Defog Agents (multi-step orchestration across SQL, Python, and R with human-in-the-loop oversight), the SQLEval evaluation framework, and MedSQL, a domain-specialized model for US healthcare billing codes (ICD-10, CPT, HCPCS). A privacy-first architecture routes only metadata (table and column names, types, descriptions) to Defog, executes queries on the customer's own infrastructure, and supports on-premise Docker deployment behind the firewall.
The technology stack pairs a Python/FastAPI backend with PostgreSQL plus pgvector for metadata and embedding storage, a React frontend, and Docker-based deployment (defog-docker-end-user, defog-backend, defog-vllm-onprem) supporting GPU (RTX4090, A10, H100) and CPU targets. Defog serves enterprise buyers through a hybrid go-to-market: direct field sales for cloud-hosted enterprise plans ($5,000/month, 20,000+ queries, SSO, white-glove onboarding) and self-hosted annual contracts, complemented by a product-led growth motion (free tier of 100 API calls/month, Python client, REST API, CLI wizard) and distribution via AWS Marketplace, GCP Marketplace, Hugging Face, and GitHub. Named enterprise customers include Toyota, Alliance Bernstein, Macmillan, and Genmab, with additional sector-specific deployments in healthcare, finance, government, and manufacturing. The company has raised approximately $2.7M across a 2023 YC-led pre-seed/SAFE and a November 2023 $2.2M seed co-led by Script Capital and Y Combinator, and achieved SOC-2 Type II compliance in October 2023.
Defog firmographics
Firmographics- Name
- Defog
- Legal name
- Full Stack Data Pte Ltd
- Website
- https://defog.ai
- Company type
- Private
- Founded year
- 2023
- Operating status
- Operating
- Headcount range
- 1–10 employees
- Short description
- Defog is a natural language data query platform that lets enterprises ask free-form questions of structured databases via its fine-tuned open-source SQLCoder LLM family. It serves enterprise customers in finance, healthcare, manufacturing, and government with privacy-first, on-premise-capable deployments.
- Ownership category
- akta.pro rank
Defog industry classification
Industry- Product category
- AI Data Analytics Platform
- NAICS
- Software Publishers (5132), Custom Computer Programming Services (541511), Computer Systems Design and Related Services (54151)
- SIC
- Services-Prepackaged Software (7372), Services-Computer Programming, Data Processing, Etc. (7370)
- akta.pro primary industry
- Data Platform (Unified Data & Analytics) Suites (HDAEABAD)
- akta.pro secondary industries
- Query Engines & SQL Analytics Layers for Warehouses/Lakes (HDAEABAI), Healthcare Data Lakes, Lakehouses & Cloud Warehouses (HLACAIAA)
Keywords
Where Defog is headquartered
LocationHeadquarters
- HQ city
- San Francisco
- HQ country
- United States
- HQ region
- North America
Offices1 record
Markets served
Defog business model
Business model- GTM type
- B2B
- Offering type
- Software
- Cost components
- Technology or R&D, Personnel, Infrastructure, Marketing or Sales, Operations
Revenue model
- Enterprise Cloud-Hosted Subscription: Enterprise tier priced at $5,000/month offering 20,000+ queries per month, one-click Docker deployment, SSO, custom AI tools, white-glove onboarding, priority support, up to 200 queries/minute, and SLA/MSA/DPA availability. Billed annually.
- Enterprise Self-Hosted Subscription: Annual commitment plan for self-hosted deployment on the organization's own infrastructure (8b, 14b, or 32b models). Includes unlimited queries and no rate limits. All Enterprise Cloud features plus on-prem hosting.
- Free Tier: Free plan with limited API access (100 queries/month), schema limited to 4 tables and 25 total columns, using a less capable model. Designed for evaluation and developer experimentation.
- Pilot Program: One-time fee of $900 for an 8-week pilot program including API access, up to 10,000 queries, metadata from up to 10 tables (100 columns), consultations with the founding team, and fine-tuning support.
Pricing tiers
| Model | Billing | Price |
|---|---|---|
| Freemium | Monthly | Free tier for evaluation and developer experimentation |
| Subscription | Annual | Enterprise Cloud-Hosted - Customizable and scalable AI data analyst for organizations |
| Subscription | Multi-year contract | Enterprise Self-Hosted - Models fine-tuned for the organization on own infrastructure |
| One time/ perpetual license | Pay-as-you-go | Pilot Program - 8-week program to evaluate Defog |
Go-to-market motion4 records
Distribution channels6 records
Marketing channels8 records
Defog product offering
Product offeringCore offering
Defog provides a natural language data query platform that converts free-form English questions into SQL queries using its proprietary SQLCoder family of fine-tuned large language models. The platform connects to enterprise databases (Snowflake, Postgres, BigQuery, Databricks, MySQL, Redshift, SQL Server, SQLite) and returns visualized answers in seconds. It is offered as a cloud-hosted SaaS subscription, a self-hosted on-premise Docker deployment, and a free developer tier, with Defog Agents extending the product to autonomous multi-step analytical workflows.
Differentiator
Problem solved
Functional benefit
Brands
- SQLCoder: Industry-leading AI model for querying structured data, developed by Defog. SQLCoder converts natural language questions into SQL queries and is available in various sizes (7B, 15B, 34B, 70B parameters).
- SQLEval
- Defog Agents
- MedSQL
Products and services
- Defog Platform An AI data analyst platform that converts natural language questions into SQL queries, executes them against enterprise databases (Snowflake, Postgres, BigQuery, Databricks, MySQL, Redshift, SQL Server, SQLite), and returns visualized results. Offered as cloud-hosted SaaS and self-hosted Docker deployment for enterprises needing on-premise data privacy.
- SQLCoder A family of open-source, fine-tuned large language models specialized for text-to-SQL conversion, available in 7B, 8B, 15B, 34B, and 70B parameter sizes. Achieves 93% accuracy on novel schemas and 99%+ when fine-tuned to a specific schema, outperforming GPT-4, Claude, and CodeLlama-70B.
- Defog Agents AI agents that decompose complex multi-step analytical questions into discrete executable tasks across SQL, Python, and R, with human-in-the-loop oversight. Automates feature engineering, model testing, and report generation in dashboards, slides, and spreadsheets.
- MedSQL A domain-specialized model for healthcare customers that supports natural language querying of US healthcare billing data using ICD-10, CPT, and HCPCS codes. Built for privacy-first deployment in healthcare environments.
- SQLEval An open-source evaluation framework that measures accuracy of LLM-generated SQL by executing both gold and generated queries against databases and comparing results, handling multiple correct SQL variants and accepting semantically equivalent outputs.
- Pilot Program An 8-week pilot program priced at $900 one-time, including API access, up to 10,000 queries, metadata from up to 10 tables/100 columns, two 1-hour consultations with the founding team, and fine-tuning support. Designed for organizations evaluating Defog before committing to an enterprise subscription.
Quantifiable outcome
- 80% reduction in analysis time; estimated savings of 2,500 hours per month for a 100-person department (publicly-listed US company)
- +5 more outcomes
Companies that use Defog
Customer profileNamed customers8 records
Segments6 records
Ideal customer profiles2 records
Defog technology and API
TechnologyTechnology focussed Yes
API detail
- Has API
- Yes
- API docs
- API detail
Core technology
AI maturity
App detail
Integration12 records
AI capability9 records
Feature12 records
Defog partnerships and signals
Strategic signalPartnerships
21 partnerships are on record, tiered core.
- SnowflakecoreSnowflake is a supported database type in Defog's API (db_type parameter). Defog can query Snowflake data warehouses in natural language, making Snowflake a core integration partner for data connectivity.
- DatabrickscoreDatabricks is supported as a database type in Defog's API and can be connected via the Python client with server hostname, access token, and HTTP path credentials.
- PostgreSQLcorePostgreSQL is natively supported as a database type. Defog's backend uses PostgreSQL with the pgvector extension for metadata and vector storage. Supports connections via SQLAlchemy ORM.
- BigQuerycoreBigQuery is supported as a database type in Defog's API. Connection is established via a service account JSON key file path.
- MySQLcoreMySQL is supported as a database type in Defog's API. Can be connected via host, database, username, and password credentials.
- Amazon RedshiftcoreAmazon Redshift is supported as a database type in Defog's API. Can be connected via host, port, database, username, and password credentials.
- SQL ServercoreMicrosoft SQL Server is supported as a database type in Defog's API. Can be connected via server, database, username, and password credentials.
- SQLitecoreSQLite is supported as a database type in Defog's API and can be used for local testing and small-scale deployments.
- Hugging FacecoreSQLCoder model weights are hosted and distributed via Hugging Face. Defog's models have received over 25,000 lifetime downloads on the platform. The Hugging Face CEO publicly endorsed SQLCoder as the best open-source model for SQL tasks.
- StarCodercoreSQLCoder and SQLCoder2 are fine-tuned implementations of the StarCoder base model from Hugging Face's BigCode project, serving as the foundational model for Defog's 7B-15B parameter text-to-SQL models.
- CodeLlamacoreSQLCoder-34B and SQLCoder-70B are fine-tuned implementations of Meta's CodeLlama model, serving as the foundational model for Defog's larger parameter text-to-SQL models.
- Mistral-7BcoreSQLCoder-7B is a fine-tuned implementation of the Mistral-7B model, enabling a smaller-footprint deployment option for text-to-SQL tasks.
- DockercoreDefog's entire deployment stack is containerized using Docker and docker-compose. Three Docker images (defog-docker-end-user, defog-backend, defog-vllm-onprem) are used for deployment. Enables automated setup in under 30 minutes.
- FastAPIcoreDefog's backend services (defog-docker-end-user and defog-backend) are built using the FastAPI Python web framework.
- SQLAlchemycoreDefog uses SQLAlchemy as an ORM for database connections, enabling support for multiple database backends including PostgreSQL, SQLServer, MySQL, Oracle, and SQLite.
- pgvectorcoreDefog uses the pgvector PostgreSQL extension to store embeddings of metadata, instructions, and golden queries, enabling fast and efficient retrieval without a separate vector database.
- ReactcoreDefog's frontend is built using React, providing the user interface for querying databases in natural language.
- PyTorchcoreDefog's LLM service (defog-vllm-onprem) uses PyTorch as the deep learning backend for running SQLCoder models.
- CUDA/cuDNNcoreDefog's LLM service image includes optimized CUDA and cuDNN drivers by default, enabling GPU-accelerated inference for text-to-SQL generation.
- vLLMcoreDefog's LLM service (defog-vllm-onprem) is built on a modified version of vLLM's Async server for handling multiple inference requests with optimized performance.
- NVIDIA (GPU)coreDefog recommends NVIDIA GPUs (RTX4090 for physical hardware, A10 for cloud, H100 for latency-sensitive deployments) for running SQLCoder models on-premises. The defog-vllm-onprem image includes CUDA drivers.
Scale indicators12 records
Recent moves6 records
Expansion highlights7 records
Defog competitors and assessment
Company assessmentDirect peers
- ThoughtSpot: AI-powered analytics platform offering natural language search and AI-driven insights on cloud data warehouses. Direct competitor to Defog in the AI data analyst category, though it is a broader incumbent with a commercial (not open-source) model.
- Hex Technologies: AI data workspace combining notebooks, SQL, Python, and natural language querying for collaborative data analysis. Overlaps directly with Defog Agents' multi-language (SQL/Python/R) analysis workflow and the AI-augmented data analyst use case.
- Mode Analytics: SQL-based analytics and BI platform with ad-hoc analysis and reporting. Direct competitor for SQL-driven data exploration workflows that Defog serves with natural language.
- Vanna AI: Open-source text-to-SQL framework that trains LLMs on enterprise schemas. Closest direct open-source peer to Defog's SQLCoder, with similar RAG-based fine-tuning approach to text-to-SQL generation.
- PandasAI: Open-source Python library enabling natural language conversations with data (pandas DataFrames and databases). Comparable to Defog in bringing NL querying to structured data, with a more developer-library focus.
Emerging players
- MindsDB: Open-source platform that brings AI/ML models directly into databases for natural language queries and predictions. Adjacent to Defog in enabling AI over enterprise data, with stronger ML-in-database heritage.
- Obviously AI: No-code AI platform for predictive analytics on tabular data. Adjacent competitor in the no-code AI data analyst space, with a focus on ML predictions rather than SQL generation.
- Numbers Station: AI agents purpose-built for enterprise data workflows including text-to-SQL, analytics, and data engineering. Closely comparable to Defog Agents' autonomous multi-step workflow capabilities.
Broad incumbents
- Snowflake Cortex: Snowflake's native AI/ML service offering built-in LLM functions including text-to-SQL across the data cloud. Competes with Defog by embedding similar capabilities directly into the most-adopted cloud data warehouse.
- Databricks Assistant / Mosaic AI: Databricks' suite of AI assistants and GenAI features for notebooks, SQL, and data engineering workflows. Competes with Defog in the enterprise data analyst category as a bundled offering within a major data platform.
Market position
Strengths5 records
Weaknesses5 records
Competitive moat5 records
Key risks6 records
Key highlights7 records
Customer concentration
Defog social profiles
Digital presenceDefog financial estimates
Financial estimateRevenue estimate
Valuation estimate
Defog leadership team
Management profileNumber of profiles
Profiles5 records
Defog funding detail
Funding detailFunding overview
Funding rounds2 records
Investors5 records
Funding detail is available on the Subscription and Enterprise plan.Contact sales →
Defog M&A and investment
M&A and investmentM&A
Investments
M&A and investment is available on the Subscription and Enterprise plan.Contact sales →
Frequently asked questions about Defog
What does Defog do?
Defog provides a natural language data query platform that converts free-form English questions into SQL queries using its proprietary SQLCoder family of fine-tuned large language models. The platform connects to enterprise databases (Snowflake, Postgres, BigQuery, Databricks, MySQL, Redshift, SQL Server, SQLite) and returns visualized answers in seconds. It is offered as a cloud-hosted SaaS subscription, a self-hosted on-premise Docker deployment, and a free developer tier, with Defog Agents extending the product to autonomous multi-step analytical workflows.
Is Defog a public or private company?
Defog is a private company. It is classified as venture growth investor backed and is currently operating.
When was Defog founded?
Defog was founded in 2023. It employs 1 to 10 people.
Where is Defog based?
Defog is headquartered in San Francisco, United States, in the North America region.
How does Defog make money?
Four revenue lines are on record. Enterprise Cloud-Hosted Subscription is the primary driver. The others are enterprise Self-Hosted Subscription, free Tier and pilot Program.
Who are Defog's main competitors?
Direct peers on record are ThoughtSpot, Hex Technologies, Mode Analytics, Vanna AI and PandasAI. Emerging players are MindsDB, Obviously AI and Numbers Station. Broad incumbents are Snowflake Cortex and Databricks Assistant / Mosaic AI.
Does Defog have an API?
Yes. Defog offers a REST API for generating SQL queries from natural language questions. The API endpoint is https://api.defog.ai/generate_query_chat and accepts parameters including api_key, db_type (supporting bigquery, databricks, mysql, postgres, redshift, snowflake, sqlite, sqlserver), question, and optional previous_context for conversation continuity. A Python client library (pip install defog) is also available which wraps the API and executes SQL queries on the user's database, returning results directly. Developer documentation is at docs.defog.ai/api-access.
What industry is Defog in?
Defog's product category is AI Data Analytics Platform. Its primary akta.pro industry code is HDAEABAD, Data Platform (Unified Data & Analytics) Suites, with a secondary code of HDAEABAI, Query Engines & SQL Analytics Layers for Warehouses/Lakes. Its NAICS code is 5132 and its SIC code is 7372.