575 Lab – Dataiku's Open Source Office

Why Open Source Matters for Dataiku

Accelerating the development of open, production-ready tools that give enterprises the transparency, control, and governance required to deploy AI with confidence.

• Trust, by design: Inspect every component of your AI stack. No black boxes, no vendor lock-in, complete visibility into how models process your data.

• Enterprise control: Deploy open-source tools to your infrastructure. Customize to your needs. Maintain security standards and data sovereignty without compromise.

• Community-Driven Innovation: With a community of developers contributing, the software benefits from diverse perspectives, rapid bug fixes, and continuous innovation. The code’s transparency also enables security scrutiny and trust.

Open source gives organizations transparent building blocks they can inspect, standardize, and govern — especially as models and agent stacks continue to evolve.

Florian Douetteau, CEO and co-founder, Dataiku

Kiji Privacy Proxy

Protect sensitive data when using closed-source LLMs.

Run a locally deployed proxy that helps teams safely use external LLM APIs by detecting and redacting sensitive data before it leaves your environment.

Built for teams that need to:

• Reduce exposure of PII and sensitive inputs/outputs

• Maintain privacy controls while still using best-in-class models

• Develop your custom PII detection models with our provided workflow

Kiji Inspector

Get full visibility into agent decision-making.

Trace and inspect multi-step agent workflows to understand what happened, why, and what to fix.

Built for teams that need to:

• Understand the business decisions performed by an agent

• Debug unexpected agent behavior

• Create an inspection trail for risk and compliance reviews

How to Contribute and Collaborate

Build with the community and ship what’s usable.

The 575 Lab is designed to work in the open: clear repos, practical tooling, and a contributor-friendly approach that helps ideas become stable, production-ready projects.

Ways to get involved:

• Join discussions, share feedback, and propose improvements

• Contribute code, docs, testing, or examples

• Help shape tools teams can actually run in production

Other Dataiku Open Source Contributions

• vllm: Dataiku’s 575 Lab is actively contributing to the most popular LLM inference platform.
• scikit-learn: We are part of the first consortium of corporate sponsors that support the development of this flagship machine learning library.
• CodeMirror: We actively sponsor this Javascript code editor.
• Cardinal: Dataiku is the author and maintainer of this Python package designed to perform and monitor active learning experiments, leveraging various query sampling methods and metrics.