2-Minute Enterprise Quickstart

This guide walks you through executing your first end-to-end analytical turn, observing sandboxed code execution, validating hallucination-resistant numerical grounding, and exporting production-ready analysis pipelines.

Step 1: Launch the Control Plane

Ensure Wizard is running via your preferred deployment channel:

# If installed via Homebrew:
wizard start
 
# If running standalone release:
./cli/wizard start
 
# If running containerized stack:
docker compose up -d

Navigate to http://localhost:3000 in your browser. The web workspace will automatically connect over WebSockets to the backend control plane (http://127.0.0.1:8000).

Step 2: Ingest Datasets & Discover Schemas

Wizard supports streaming ingestion for structured and semi-structured formats without loading entire multi-gigabyte files into web browser memory:

1. Drag-and-Drop Ingestion: Drop CSV, Parquet, JSON/NDJSON, Excel (.xlsx), Feather, or TSV files directly into the workspace canvas.
2. Relational Database Connectors: Alternatively, click Add Connection to mount PostgreSQL, MySQL, SQLite, DuckDB, ClickHouse, or Snowflake data warehouses.
3. Multi-Table Namespace: Every ingested dataset is automatically indexed into the session catalog and exposed to generated Python scripts as:
   python

   Copy

   # Addressable table dictionary
   df = tables["sales_2026"]
   customers = tables["dim_customers"]

Step 3: Issue Natural-Language Analytical Inquiries

Unlike basic SQL-generation wrappers, Wizard uses an iterative Observe -> Decide -> Act loop designed for open-ended, multi-step business inquiries:

"Analyze our Q3 enterprise churn. Which customer segments experienced the highest contract downgrades, what are the primary statistical drivers, and how does discounting correlate with churn probability?"

Selecting Execution Tier / Depth

Select your desired analytical depth from the composer toolbar:

• Fast Tier: 1 iteration pass, immediate exploratory statistics.
• Balanced Tier (Default): Multi-step investigation with schema probing, dynamic chart generation, and hypothesis validation.
• Deep Research Tier: Full multi-turn investigation with automated competitive hypothesis testing, The Council adversarial review, and independent dual-route verification.

Step 4: Inspect the Streaming Execution Cycle

As the agent deliberates, the live UI streams each discrete step in real time:

[PLAN]      Manager forms 3-stage hypothesis on contract discounts vs churn.
[CODE]      Worker writes vectorized pandas & scikit-learn logistic regression.
[GUARD]     CodeGuard AST scan: 0 policy violations detected (safe).
[SANDBOX]   Host Subprocess executed (124ms) — stdout captured.
[SYNTHESIS] Manager synthesizes statistical findings with numerical citations.

If generated code triggers an unexpected exception (e.g. key mismatch or timestamp parsing error), Wizard's Self-Correction Engine catches the traceback, isolates the failure reason, and generates a corrected AST script without crashing the session.

Step 5: Validate Trust & Numerical Provenance

Enterprise compliance requires verifiable truth. Wizard implements three distinct trust gates on every analytical response:

1. Strict Provenance Pinning: Every integer, float, and percentage cited in the final synthesis is traced to raw stdout / DataFrame outputs. Unreferenced numbers trigger a red warning badge.
2. Dual-Calculation Verification: The core metric (e.g. Churn Rate = 14.2%) is independently recalculated through an alternate mathematical route to eliminate arithmetic artifacts.
3. Implicit Assumption Auditing: The side panel surfaces all implicit data transformations (dropped NaN rows, coerced date formats, filtered outliers, or join cardinality choices).

Step 6: Export & Automate

Once satisfied with the findings:

• Export Reproducible Python Script: Download a fully self-contained .py script that can be scheduled via Airflow, Prefect, or cron.
• Export Executive PDF / Markdown Report: Generate a formatted evidence-backed report with embedded interactive charts.
• Promote to Skill: Save the analytical pattern as a reusable corporate skill via wizard skills for your team.

Next Steps

• Explore the complete CLI Reference Guide for automated scripting.
• Review Data Modes & Air-Gap Privacy before connecting production databases.
• Read Execution Sandboxing & CodeGuard to understand kernel isolation boundaries.