Kubernetes-native workflow runtime for batch DAGs and streaming pipelines.

---

BubuStack is an open-source toolkit for declarative, composable workflows on Kubernetes. Define workflows as CRDs, review and promote them through pull requests, and run them with durable execution semantics for batch automation and streaming workloads.

Why BubuStack

• GitOps-native - Workflows are CRDs. Version, review, and promote them through pull requests like any other infrastructure.
• Durable execution - StoryRun and StepRun resources give you retries, timeouts, status, and observability at the workflow level.
• Composable building blocks - Reusable Engrams (tasks) and Impulses (triggers) let you package common behavior without coupling everything into one controller.
• Batch + streaming in one model - Use the same Story abstraction for DAG-style batch jobs and transport-backed streaming pipelines.

Quickstart

For fresh cluster setup, cert-manager, storage, Helm installation, and the first example, start with the website quickstart:

• Quickstart
• Prerequisites
• Artifact Hub: bobrapet Helm chart
• Artifact Hub: bobravoz-grpc Helm chart

Repositories

Repository	Description
bobrapet	Story, StoryRun, Engram, and Impulse controllers — the core operator
core	Shared contracts, templating engine, transport runtime, and identity helpers
bobravoz-grpc	Transport operator for streaming workloads with hub and peer-to-peer topologies
bubu-sdk-go	Go SDK for building Engrams and Impulses with streaming, retries, and observability
tractatus	Canonical protobuf definitions and multi-language transport artifacts
examples	Runnable examples for batch and streaming workflows
website	Public documentation site and product pages
bubuilder	UI and API work for browsing runs, stories, and engrams
bubu-registry	Registry and packaging work for sharing Engrams and Impulses
helm-charts	Helm charts for bobrapet, bobravoz-grpc, and bubuilder
engrams/*	First-party Engrams — HTTP, OpenAI, LiveKit Bridge, MCP, VAD, Map-Reduce, and more
impulses/*	First-party Impulses — Cron, GitHub webhooks, Kubernetes events, LiveKit webhooks

Examples

Example	Pattern	What it shows
Hello World	Batch	Zero-dependency quickstart - parallel steps, DAG dependencies, template expressions
Scheduled Health Check	Batch	Cron triggers, parallel fan-out, allowFailure, AI-powered summarization
Map-Reduce Summarizer	Batch	Dynamic parallelism with sub-stories, map-reduce adapter, result aggregation
Content Digest (Simple)	Batch	RSS to AI extraction to newsletter to Discord via MCP adapter
Content Digest (Advanced)	Batch	Multi-feed fan-out with map-reduce, combined digest, Discord notification
Multi-Model Consensus	Batch	Same template with different configs, fan-in judge pattern, structured output
GitHub PR Review	Batch	AI-powered PR review - GitHub webhooks, diff analysis, automated comments
Materialize Demo	Batch	Offloaded data handling - storage refs, materialization, controller-mode resolution
Pod Crash Notifier	Batch	Kubernetes event watch, AI crash analysis, Discord alerts
LiveKit Text Chat	Streaming	Text-only chat assistant via LiveKit data channels
LiveKit Live Translation	Streaming	English-to-French speech translation with VAD, STT, LLM, and TTS
LiveKit Voice Assistant	Streaming	Voice and chat dual-mode - VAD, STT, LLM, TTS with a parallel chat path

Watch the LiveKit realtime examples demo: YouTube.

Contributing

We welcome contributions of all kinds — bug fixes, new Engrams, documentation, and feature proposals.

• Star the repos and watch releases
• Build an Engram or Impulse using the Go SDK
• Report bugs and request features in the affected repository
• Pick up a good first issue or help wanted
• Discuss architecture and roadmap in Discussions

Read the full Contributing Guide to get started.

Community

• Documentation — Guides, API reference, and architecture deep dives
• GitHub Discussions — Questions, proposals, and community conversations
• Discord — Real-time chat for contributors and users
• Security Policy — Responsible disclosure process
• Code of Conduct — Contributor Covenant 3.0

---

_{Apache 2.0 Licensed. Built for platform engineers, AI/ML teams, and anyone who believes workflows belong in Git.}