⚡ fds — Fabric Data Science CLI

Your terminal is your workspace. One CLI for the entire Fabric DS lifecycle — from experiment to endpoint.

fds is a command-line tool for Microsoft Fabric Data Science that lets you manage experiments, models, notebooks, environments, and endpoints without ever opening a browser. Designed for data scientists who live in the terminal and AI agents that need structured, scriptable output.

# Train → Track → Deploy in three commands
fds experiment runs list my-exp --sort-by metrics.auc --limit 1 -o json
fds model register churn-model --from-run <run-id>
fds endpoint activate churn-model --version 1 --wait

---

✨ Key Features

Feature	Description
🧪 Full ML Lifecycle	Experiments → Runs → Models → Endpoints — all from the terminal
🤖 Agent-First Design	Every command supports --help for self-discovery and --output json for structured parsing
📓 Notebook Ops	Upload, push, run (with -p key=value or JSON parameters), and download notebooks
🌍 Environment Management	Create environments, manage pip/conda libraries, publish — no portal clicks
🏗️ Lakehouse & Spark	Upload files, import data, run Spark jobs, vacuum Delta tables
📊 DAX Queries	Execute DAX queries against semantic models directly from CLI
🔐 Flexible Auth	Browser, Azure CLI, device-code, and service-principal authentication
🎯 Structured Output	table (human), json (agent), csv/tsv (export) — every command, every format
⏱️ Async & Wait	Long-running operations (notebook run, env publish) support --wait and --timeout
💡 Actionable Errors	Every error includes a hint: what went wrong and what to do next

---

🚀 Quick Start

Install

pip install fabric-ds-cli

Authenticate & Configure

fds auth login                              # browser-based login
fds config set workspace "My-Workspace"     # set default workspace
fds auth status                             # verify connection

Your First Workflow

# 1. List experiments
fds experiment list

# 2. View recent runs, sorted by accuracy
fds experiment runs list my-experiment --since 7d --sort-by metrics.accuracy

# 3. Compare top runs side by side
fds experiment runs compare --experiment my-experiment --last 3

# 4. Register the best model
fds model register churn-model --from-run <best-run-id>

# 5. Deploy to real-time endpoint
fds endpoint activate churn-model --version 1 --wait

# 6. Score!
fds endpoint score churn-model --data '{"inputs": [[1, 2, 3]], "formatType": "dataframe", "orientation": "values"}'

Explore Semantic Models

# List semantic models in the workspace
fds dataset list

# Inspect tables and measures
fds dataset tables sales-model
fds dataset measures sales-model

# Run DAX queries directly from the terminal
fds dataset dax "EVALUATE TOPN(10, 'Sales', 'Sales'[Amount], DESC)" --dataset sales-model

---

🤖 For AI Agents

Install Skills (Copilot / Claude / Cursor)

fds ships pre-built Skills that teach agents how to use the CLI:

npx skills add <fds-repo> --skill ds-authoring-cli     # ML lifecycle (create, deploy, score)
npx skills add <fds-repo> --skill ds-consumption-cli    # Read-only (query, compare, explore)

Manual: cp -r .github/skills/ds-* ~/.copilot/skills/

Other Frameworks

For LangChain, AutoGPT, OpenAI Agents SDK, or custom orchestrators — see the Agent Integration Guide (output contract, exit codes, system prompt template, retry patterns, headless auth).

---

📖 Command Reference

📘 Full reference with all parameters and examples: docs/commands.md

Core Workflow

Command	Subcommands	Description
fds experiment	list · create · get · delete	ML experiment management
fds experiment runs	list · get · compare	Run tracking, comparison, filtering
fds model	list · get · versions · register · delete	Model registry operations
fds endpoint	list · get · versions · activate · deactivate · score · update	Real-time endpoint lifecycle

Infrastructure

Command	Subcommands	Description
fds notebook	list · upload · push · download · delete	Notebook management
fds notebook run	start · status · cancel · list	Notebook execution & monitoring
fds env	list · create · get · add-libraries · remove-library · publish · delete	Environment & dependency management
fds spark-job	list · run · status · cancel	Spark job definitions
fds lakehouse	create · get · delete · list · tables · vacuum · upload · import	Lakehouse data operations
fds dataset	list · tables · measures · refresh · dax	Semantic models & DAX queries

System

Command	Subcommands	Description
fds auth	login · status · logout	Authentication management
fds config	set · get · list · reset	CLI configuration
fds workspace	create · delete · list · items · set	Workspace management

💡 Every command supports --help for detailed usage, arguments, and examples.

Global Options

fds [command] --output json     # Output format: table | json | csv | tsv
fds [command] --workspace "X"   # Override default workspace
fds [command] --verbose         # Debug-level logging
fds [command] --quiet           # Suppress non-essential output

---

🏗️ Architecture

┌──────────────────────────────────────────────────────────┐
│                    fds CLI (Typer)                        │
│  fds experiment · model · notebook · endpoint · env ...  │
├──────────────┬───────────────────────┬───────────────────┤
│  commands/   │     services/         │     core/         │
│  (CLI layer) │  (business logic)     │  (infrastructure) │
│  parse args  │  one service per      │  config, errors,  │
│  → service   │  resource type        │  auth, output,    │
│  → render    │                       │  workspace        │
├──────────────┴───────────┬───────────┴───────────────────┤
│                          │                               │
│    MLflow Tracking API   │   Fabric REST API             │
│    (experiment, model,   │   (notebook, env, endpoint,   │
│     run, metric, tag)    │    lakehouse, spark-job)       │
│                          │                               │
├──────────────────────────┴───────────────────────────────┤
│              Microsoft Fabric Platform                    │
└──────────────────────────────────────────────────────────┘

Design principles:

• Commands are thin: parse CLI args → call service → render output
• Services own business logic: one class per Fabric resource type
• Core handles cross-cutting concerns: auth, config, errors, formatting
• No raw HTTP — all platform calls go through established client libraries

---

⚖️ fds vs fab CLI

Capability	fds	fab
ML experiments & runs	✅ Full CRUD + compare	❌
Model registry	✅ Register, version, delete	❌
Real-time endpoints	✅ Activate, score, scale	❌
Notebook execution	✅ Run with params, poll status	⚠️ Basic
Environment management	✅ Libraries, publish	❌
DAX queries	✅	❌
Workspace file operations	⚠️ Lakehouse upload	✅ Full (ls, cd, cp, rm)
Git integration	❌	✅
Permissions management	❌	✅
AI Agent optimized output	✅ --output json	❌

TL;DR: fds owns the DS lifecycle (experiment → model → endpoint). fab owns platform infrastructure (files, Git, permissions). They complement each other.

---

🔄 CI/CD

# GitHub Actions example
steps:
  - run: pip install fabric-ds-cli
  - run: fds auth login -m service-principal --tenant-id $TENANT --client-id $CLIENT_ID --client-secret $SECRET
  - run: |
      fds notebook run start training \
        -p version="$GITHUB_SHA" --wait
  - run: |
      BEST=$(fds experiment runs list my-exp \
        --sort-by metrics.auc --limit 1 -o json | jq -r '.[0].run_id')
      fds model register prod-model --from-run "$BEST"
      fds endpoint activate prod-model --version 1 --wait

---

🔧 Prerequisites

• Python 3.10+
• Microsoft Fabric workspace with appropriate permissions
• Azure CLI — az login for interactive authentication

Optional

• .NET SDK 8.0 — required only for fds dataset dax and fds dataset measures (PythonNet bridge)

  # macOS
  brew install dotnet-sdk
  # Ubuntu/Debian
  sudo apt install dotnet-sdk-8.0
  # Windows — https://dotnet.microsoft.com/download
  

  If .NET is not installed, DAX/measures commands show installation instructions instead of crashing.

---

🛠️ Development

Setup

git clone https://github.com/microsoft/fabric-ds-cli.git
cd fabric-ds-cli
uv venv .venv --python 3.12
source .venv/bin/activate
uv pip install -e ".[dev]"

Note: Always use uv for dependency management. Bare pip may resolve to the system Python.

Testing

pytest tests/ -v                          # All unit tests
pytest tests/ -k "not fabric" -v          # Unit only (no Fabric auth needed)
pytest tests/ -m fabric -v                # E2E tests (requires Fabric workspace)

E2E tests require az login and a configured workspace. See Testing Guide for full setup.

Linting & Type Checking

ruff check src/ tests/                    # Lint (E, F, W, I, UP, B, SIM, D, C901)
mypy src/fabric_ds_cli/                   # Type check (strict mode)

Project Structure

src/fabric_ds_cli/
├── main.py              # CLI entry point, command registration
├── core/                # Infrastructure (config, errors, auth, output)
├── commands/            # Thin CLI layer (11 command groups)
└── services/            # Business logic (one service per resource)

---

📄 License

MIT