Open-source Julius AI alternatives

DB-GPT is an open-source AI data assistant for the next generation of products that combine AI and data. It connects to databases, CSV and Excel files, data warehouses and knowledge bases, then lets people ask questions in natural language while the AI writes SQL and code to answer them.

Beyond question answering, it supports agentic data analysis: planning a task, breaking it into steps, calling tools and finishing the analysis end to end. It can run Python and code-driven workflows, load reusable skills for domain-specific tasks, and generate charts, dashboards, HTML reports and summaries. Tasks run in sandboxed environments for safety. Beyond the assistant, it serves as a platform for building AI-native data agents and applications with its AWEL workflow language, retrieval-augmented generation and multi-model support, with topics naming models such as DeepSeek and Vicuna.

It is written in Python and licensed under MIT, with documentation, a community and a research paper. Privacy and security are highlighted in the topics, and it can run on your own infrastructure with local models. It suits data teams that want natural-language access to databases while keeping data in-house.

DeepAnalyze is an agentic large language model for autonomous data science, developed by researchers at Renmin University of China and Tsinghua University. The goal is for it to complete data-centered tasks without step-by-step human guidance, acting as an AI data analyst that takes in large amounts of data and produces a professional analysis report.

According to the README it can run the whole data science pipeline, including data preparation, analysis, modeling, visualization and report generation. It also supports open-ended data research across different data sources, from structured data such as databases, CSV files and Excel sheets to other formats. Repository topics mention Qwen, Jupyter, deep research and data visualization, and the code is Python.

DeepAnalyze is released under the MIT license. It is a research-oriented project you run yourself with the provided model and code, rather than a hosted product. It suits data scientists, analysts and researchers exploring autonomous analysis agents.

nao is an open-source framework for building an analytics agent and deploying it as a chat interface. Data teams assemble the agent's context with the nao-core command-line tool, then give business users a UI where they can ask questions in natural language and receive analysis, charts and the reasoning behind each answer.

On the data team side, the context builder works like a file system and can include data, metadata, documentation, tools and MCP servers. The framework is described as stack-agnostic, with topics pointing to BigQuery, Snowflake, Databricks and PostgreSQL. Teams can unit test the agent's performance before release, version the context, track usage and collect feedback from users who mark answers right or wrong.

nao is written in TypeScript and is designed to be self-hosted with your own LLM keys, which keeps data under your control. A Slack connection is available as an option during setup. It suits data and analytics engineering teams who want a governed text-to-SQL assistant for the rest of the company rather than ad hoc prompts.

Pretzel is a fork of Jupyter aimed at data scientists and analysts who want the familiar notebook experience enhanced with AI assistance. Because it is built directly on Jupyter, existing configuration, settings, keybindings and extensions carry over automatically when switching from plain Jupyter.

On top of standard notebooks, Pretzel adds inline tab completion as you type in a cell, an 'Ask AI' prompt triggered from a cell or a keyboard shortcut, and an AI sidebar for chatting, generating code and asking questions, with an @ trigger for autocompleting function and variable names. Users can bring their own model from OpenAI, Anthropic/Claude, Ollama or Groq. The project's roadmap includes native AI code generation similar to Cursor, realtime collaboration features, SQL support inside notebooks, a visual analysis builder, and one-click dashboard creation from notebooks.

Pretzel is written in TypeScript and installed with pip, after which the pretzel lab command opens the web interface; a free hosted version is also available at pretzelai.app. The repository metadata lists its license as Other, so check the license file for exact terms.