ade-ops
Most open-source data platform tools hand you a sample and walk away.
ade-ops asks what you have.
An open-source framework for analytics teams working across Databricks, Microsoft Fabric, and Power BI. Point it at a single project to start — the reference distribution isn't a static demo, it's a scenario-aware starter kit that asks about your environment and scaffolds what fits — then scale to how your whole team operates, with people and AI agents in control.
Status: F1 public preview (early). Apache License Version 2.0. Maintained by Roberto Butinar.
What ade-ops does
- Single source of truth in
src/ with overlay-based transforms for multi-environment deployment. Author once; the same engine treats Databricks, Fabric and Power BI the same way.
- Remote workspaces are authoritative. Promotions across environments are visible — diff before push, no silent overrides.
- The framework admits what it doesn't know. Skills carry a
status: field (experimental / preview / stable) so you can calibrate trust. Notebook converters tag output as compat, light, heavy, or impossible — the last two are your call, not the tool's.
- People and AI agents, in control. ade-ops is usable from a plain CLI and reaches its potential when paired with an AI coding assistant that can drive its skills — with a human confirmation on anything that touches a live environment.
Three onboarding scenarios in V1
When you run the onboarding skill, ade-ops asks which scenario fits your environment:
| Scenario | When to pick it | Typical user |
|---|
databricks-to-powerbi (default) | You have Databricks + Power BI (Pro/Premium). You want to manage notebooks + PBIR reports + semantic models end-to-end. | Most enterprise teams |
databricks-to-fabric | You have Databricks + a Microsoft Fabric tenant and want the full medallion → Fabric lakehouse → DirectLake semantic model + Power BI report chain. | Teams targeting Fabric / DirectLake |
databricks-only | You have only Databricks. No Fabric / no Power BI for now. You want notebook + job deployment per environment. | Pre-PBI teams, data-eng-focused teams |
All scenarios are BYO: you bring your workspace, your notebooks, and your identity. ade-ops scaffolds the workflow.
Future scenarios (fabric-only, playground-zero-setup with synthetic sample data) are tracked for V2/V3.
Quick start
ade-ops is agent-driven in two phases: a small terminal bootstrap, then
onboarding driven by an AI coding assistant.
git clone https://github.com/rbutinar/ade-ops.git
cd ade-ops
python -m venv .venv && source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install -r requirements.txt
npm i -g @anthropic-ai/claude-code # needs Node.js LTS
claude # launch from inside the repo
Then, inside the Claude Code session, run the canonical entry point — it asks
which scenario fits and routes you to the matching setup:
/ade-ops-onboarding
→ Full walkthrough (prerequisites, credentials, verify, your first
operations): docs/getting-started/ ·
per-scenario manual steps: docs/quickstart/.
Architecture
ade-ops separates the framework core from per-deployment distributions.
ade-ops/
├── core/ # Generic, client-agnostic
│ ├── engine/ # config / overlay / state / operations
│ ├── connectors/ # Databricks, Fabric, Power BI
│ ├── conventions/ # naming, sanitization, status protocol
│ ├── playbooks/ # operational patterns
│ └── templates/ # scaffolding
│
└── distributions/ # Per-deployment customization
└── reference/ # The public reference implementation
└── projects/<project>/ # Self-contained: src + overlays + config
A distribution consumes core/ without modifying it. Each project under a distribution is self-contained — its own config, credentials, overlays, and state.
The assembly pipeline (src/ + overlays/{env}.yaml + patches/{env}/ → deployed) is the contract:
Two deployment shapes
ade-ops supports two operational shapes:
- Community / self-service (this repo): you clone, run onboarding, operate your own instance. Updates arrive via
git pull from upstream. Feedback flows via GitHub Issues.
- Enterprise (not in this repo): a private lab managed by a framework operator, with a shared DevOps mirror and team-member sandboxes. Suitable for large governance-heavy teams. If you're investigating an enterprise deployment, open an issue tagged
enterprise-inquiry or reach the maintainer directly.
Documentation
Full docs live in docs/:
Reference for AI coding assistants: CLAUDE.md (Claude Code) ·
AGENTS.md (Copilot / Codex / generic) · operational patterns in
core/playbooks/, conventions in core/conventions/.
This is the beginning of the README. Read the rest at the original. Read the full README ↗
