Denodo AI SDK

The Denodo AI SDK is an open-source component designed to streamline the integration of the Denodo Platform with Large Language Models (LLMs). It provides developers with the essential tools to build high-performance AI agents that can interact natively with enterprise data through a governed data virtualization layer.

By automating the complexities of Retrieval-Augmented Generation (RAG) and VQL query generation, the SDK ensures AI responses are grounded in real-time, factual enterprise context.

Key features

• Text-to-VQL: Automatically translates natural language questions into Denodo Virtual Query Language (VQL), allowing AI agents to query the Denodo Platform directly without manual SQL/VQL coding.
• Metadata Search (RAG): Uses vectorization to index and search through your technical and business metadata. This enables agents to find the right data assets instantly using semantic search.
• DeepQuery Agent: A sophisticated research agent that orchestrates complex tasks. It crafts a multi-step execution plan and carries it out by combining metadata discovery with live VQL data extraction.
• Model & Vector Store Agnostic: Provides a flexible architecture to configure and switch between various LLMs and Vector Databases depending on your performance and privacy requirements.

To showcase the AI SDK’s capabilities, a sample chatbot application is included.

The complete user manual for the Denodo AI SDK is available here.

Installation

To get started with the AI SDK:

1. Clone this repository and cd into it
2. Create a new virtual environment (python -m venv venv) in the root of the AI SDK's path
3. Activate the virtual environment (source venv/bin/activate for Linux/MacOS or .\venv\Scripts\activate for Windows)
4. Install the requirements.txt (python -m pip install -r requirements.txt)
5. Rename the configuration templates for both AI SDK (api/utils/sdk_config.env.example => api/utils/sdk_config.env) and the sample chatbot (sample_chatbot/chatbot_config.env.example => sample_chatbot/chatbot_config.env)
6. Review the configuration files for both the AI SDK and the sample chatbot and configure your own LLM/embeddings providers

AI SDK Benchmarks

We test our text-to-VQL pipeline on our propietary benchmark across the whole range of LLMs that we support. The benchmark dataset consists of 50+ questions in the finance sector. You may use this benchmark as reference to choose an LLM model.

Latest update: 05/18/2026 on AI SDK version 1.2

LLM Provider	Model	🎯 Accuracy	🔢 Input Tokens	🔡 Output Tokens	📊 Total Tokens	💰 Cost per Query
OpenAI	gpt-5.5	🟢	7,092	396	7,488	$0.047
OpenAI	gpt-5.5-pro	🟢	4,335	2,330	6,665	$0.549
OpenAI	gpt-5.4	🟢	7,825	433	8,258	$0.026
OpenAI	gpt-5.4-pro	🟢	3,967	1,643	5,610	$0.415
OpenAI	gpt-5.4-nano	🟡	8,632	509	9,141	$0.002
OpenAI	gpt-5.4-mini	🟢	6,942	394	7,336	$0.007
OpenAI	gpt-oss-120b	🟡	6,636	947	7,583	$0.001
OpenAI	gpt-oss-20b	🟢	7,098	2,172	9,270	$0.001
OpenAI	gpt-5.3-codex	🟡	7,601	446	8,047	$0.020
Google	gemini-3.1-pro-preview	🟢	7,825	3,726	11,551	$0.060
Google	gemini-3-flash-preview	🟢	9,397	456	9,853	$0.006
Google	gemini-3.1-flash-lite-preview	🟢	8,430	442	8,872	$0.003
Google	gemma-4-31b-it	🟢	8,527	442	8,969	$0.001
Google	gemma-4-26b-a4b-it	🟡	9,103	423	9,526	$0.001
Anthropic	claude-opus-4.7	🟢	12,994	501	13,495	$0.077
Anthropic	claude-sonnet-4.6	🟢	8,770	424	9,194	$0.033
Anthropic	claude-haiku-4.5	🟢	8,856	469	9,325	$0.011
DeepSeek	deepseek-v4-pro	🟡	8,856	481	9,337	$0.004
DeepSeek	deepseek-v4-flash	🟡	8,492	402	8,894	$0.001
MoonshotAI	kimi-k2.6	🟡	8,068	691	8,759	$0.008
Qwen	qwen3.6-plus	🟡	7,795	406	8,201	$0.003
Qwen	qwen3.6-27b	🟡	8,215	587	8,802	$0.005
Qwen	qwen3.6-35b-a3b	🟢	8,356	763	9,119	$0.002
Qwen	qwen3.6-max-preview	🟡	8,247	476	8,723	$0.012
Qwen	qwen3.6-flash	🟡	8,238	774	9,012	$0.002
xAI	grok-4.3	🔴	8,358	345	8,703	$0.011
MiniMax	minimax-m2.7	🟡	6,915	3,080	9,995	$0.006
MistralAI	mistral-small-2603	🟡	8,637	457	9,094	$0.002
MistralAI	ministral-14b-2512	🟡	9,251	677	9,928	$0.002
MistralAI	mistral-medium-3-5	🟡	8,889	427	9,316	$0.017
MistralAI	devstral-2512	🟡	8,992	424	9,416	$0.004
Meta	llama-4-maverick	🟡	7,620	489	8,109	$0.001
Meta	llama-4-scout	🟡	9,469	718	10,187	$0.001
Z-AI	glm-5.1	🟡	8,724	442	9,166	$0.001
Z-AI	glm-5-turbo	🟡	8,063	360	8,423	$0.011

Important things to note:

• "Input Tokens", "Output Tokens", and "Total Tokens" are the average per query.
• Any model with its size in the name, i.e.: gpt-oss-20b, represents an open-source model.
• Each color corresponds to the following range in terms of accuracy:
  • 🟢 = 90%+
  • 🟡 = 80%–89%
  • 🔴 = <80%

List of supported LLM providers

The Denodo AI SDK supports the following LLM providers:

• OpenAI
• AzureOpenAI
• Bedrock
• Google
• GoogleAIStudio
• Anthropic
• NVIDIA
• Groq
• Ollama
• Mistral
• SambaNova
• OpenRouter

Where Bedrock refers to AWS Bedrock, NVIDIA refers to NVIDIA NIM and Google refers to Google Vertex AI.

List of supported embedding providers + recommended

• OpenAI (text-embedding-3-large)
• AzureOpenAI (text-embedding-3-large)
• Bedrock (amazon.titan-embed-text-v2:0)
• Google (text-multilingual-embedding-002)
• Ollama (qwen3-embedding:8b)
• Mistral (mistral-embed)
• NVIDIA (baai/bge-m3)
• GoogleAIStudio (gemini-embedding-exp-03-07)

Where Bedrock refers to AWS Bedrock, NVIDIA refers to NVIDIA NIM and Google refers to Google Vertex AI.

Licensing

Please see the file called LICENSE.