Kionga documentation

Your platform. Your infrastructure.

Machine learning, data and AI agents. One platform to build, train, serve and observe models and AI agents: workspaces, pipelines, a model registry, feature stores and agent operations behind a single console and API, running in your own environment.

Start building See how it works

Your first project, from notebook to production.

Start with the installation guide, then follow the quickstart. Need a specific answer? Use search or open Ask Kionga.

Find your starting point

• Install Kionga

  Set up your environment on a laptop, a server or Kubernetes.

  Installation guide

• Build your first project

  Follow the quickstart from workspace to a running model.

  Follow the quickstart

• Keep up with changes

  Read versioned documentation updates and release notes.

  Read the changelog

Explore the platform

• Workspaces

  ---

  Personal JupyterLab and browser IDE workspaces with Git, project templates and resource limits set by your administrators.

  Workspaces

• Pipelines

  ---

  Define container DAGs once and run them with parameters, retries, timeouts, resource limits and a live run view.

  Pipelines

• Models and serving

  ---

  Track experiments, register models, promote them through quality gates and serve them with canary releases and rollback.

  Models and serving

• Features

  ---

  Online and offline features with a Feast-compatible API, or connect the feature store you already run.

  Features

• AI agents

  ---

  Deploy agents with persistent sessions, tools, tracing and token cost accounting, built with LangGraph or your own framework.

  Agents

• Real-time processing

  ---

  Stream events through Kafka to online features and live models or agents, with runnable fraud, call-center and recommendation examples.

  Real-time processing

Built for teams that run their own platform

• Secure by default

  ---

  Single sign-on, role and policy based access, deny-by-default quotas and an immutable audit trail.

  Access control

• Your infrastructure

  ---

  Run on a laptop, a single server or managed Kubernetes on AWS, Google Cloud or Azure. Your data stays in your environment.

  Deployment options

• Open standards

  ---

  Built on MLflow, Prefect, Kafka, PostgreSQL and S3-compatible storage, with an OpenAI-compatible LLM gateway. MIT licensed.

  Open source

Start here

• New to Kionga?

  Install the platform on your machine and take your first project from notebook to served model.

  Install Kionga

• Evaluating for your team?

  Understand the concepts, the architecture and what a production deployment requires.

  What is Kionga

• Integrating with Kionga?

  Use the REST API, the CLI or the Python SDK from your own tools.

  Reference