Proc. VLDB Endow. 2025 · Vol. 18, No. 12 · Beyond Quacking

What is Flock?

Flock is a DuckDB extension that deeply integrates LLM and RAG capabilities into OLAP systems. Mix analytics with semantic analysis in SQL using map and reduce functions for generation, classification, embeddings, and more. It runs on native platforms and in the browser via DuckDB-WASM.

Integrating LM and IR capabilities in SQL

Flock integrates advanced language models into RDBMSs, enabling powerful tasks like text generation, classification, filtering, and embeddings directly through SQL queries. Streamline your data workflows with AI-driven insights.

Administrative Views for Managing Prompts and Models

Treat models and prompts as resources similar to tables. You can CREATE PROMPT(s) and MODEL(s) and iterate on them without making changes to SQL queries.

Scalar / Map Functions for Tuple and Passage Transformation

Map tuples or text passages to new text, boolean, or JSON values using LLMs to support various types of semantic analysis such as extraction and classification.

Aggregate / Reduce Functions to Reduce over Tuples and Passages

Use Aggregate / Reduce functions to reduce tuples or text passages using LLMs to support various types of semantic analysis such as reranking and summarization.

-- 1) Create a model
CREATE MODEL(
  'product_summarizer',
  'gpt-4o',
  'openai'
);

-- 2) Call it from SQL
SELECT
  product_id,
  name,
  llm_complete(
    {'model_name': 'product_summarizer'},
    {
      'prompt': 'Summarize this product in one sentence.',
      'context_columns': [{'data': name}]
    }
  ) AS short_description
FROM products
LIMIT 3;

ID	Product	Short Description (from LLM)
1	Wireless Headphones	Comfortable Bluetooth headphones with clear sound and long battery life.
2	Gaming Laptop	High-performance laptop designed for modern games and creative workloads.
3	Smart Watch	Everyday smartwatch that tracks activity, notifications, and heart rate.

Why use Flock?

Create reusable models and prompts once, then call them from simple SQL.
Run LLM workloads next to your data without custom services or glue code.
Keep results in tables you can join, filter, and visualize with regular DuckDB queries.

See Installation to get started, then Getting Started for provider setup. Browse Scalar and Aggregate functions for the full API.

Paper

Beyond Quacking: Deep Integration of Language Models and RAG into DuckDB Anas Dorbani, Sunny Yasser, Jimmy Lin, Amine Mhedhbi Proc. VLDB Endow. 2025, Vol. 18, No. 12 If you found Flock useful, cite us! https://doi.org/10.14778/3750601.3750685

@article{10.14778/3750601.3750685,
  author  = {Dorbani, Anas and Yasser, Sunny and Lin, Jimmy and Mhedhbi, Amine},
  title   = {Beyond Quacking: Deep Integration of Language Models and RAG into DuckDB},
  journal = {Proc. VLDB Endow.},
  year    = {2025},
  volume  = {18},
  number  = {12},
  doi     = {10.14778/3750601.3750685},
  url     = {https://doi.org/10.14778/3750601.3750685}
}

Home

Getting Started

Resource Management

Scalar Functions

Aggregate Functions

Hybrid Search

Operations

Contributing

What is Flock?

What is Flock?

Integrating LM and IR capabilities in SQL

Administrative Views for Managing Prompts and Models

Scalar / Map Functions for Tuple and Passage Transformation

Aggregate / Reduce Functions to Reduce over Tuples and Passages

Why use Flock?

Paper

​What is Flock?

​Integrating LM and IR capabilities in SQL

Administrative Views for Managing Prompts and Models

Scalar / Map Functions for Tuple and Passage Transformation

Aggregate / Reduce Functions to Reduce over Tuples and Passages

​Why use Flock?

​Paper

What is Flock?

Integrating LM and IR capabilities in SQL

Why use Flock?

Paper