io.github.benseverndev-oss/goldenflow

Official

by benseverndev-oss · Python

GoldenFlow

Standardize, reshape, and normalize messy data — CSV, Excel, Parquet, S3, databases.

Clean and reshape messy data from multiple sources and formats.

data normalizationetlcsvexcelparquets3data cleaningdata transformation

Topics

clidata-cleaningdata-engineeringdata-qualitydata-transformationetlpolarspythonmcpa2aagentgolden-suitemcp-serverdata-pipelinedata-normalizationedge-computingjavascriptnodejsnpmtypescript

README.md26 KBView on GitHub

Moved. This repo has moved into the benzsevern/goldenmatch monorepo at packages/python/goldenflow (and packages/typescript/goldenflow)/. This repo is archived; new development happens in the monorepo.

GoldenFlow

Data transformation toolkit — standardize, reshape, and normalize messy data before it hits your pipeline. Built by Ben Severn.

Works on files (CSV, Excel, Parquet), cloud storage (S3, GCS), or live databases. Zero-config mode auto-detects what needs fixing. One command to clean what GoldenCheck found and prep what GoldenMatch needs. Available in Python and TypeScript with full feature parity.

bash

# Python
pip install goldenflow
goldenflow transform data.csv

# TypeScript / Node.js
npm install goldenflow
npx goldenflow-js transform data.csv

The Problem

Your data arrives broken in predictable ways:

Phone numbers come in 15 different formats
Dates are mixed between MM/DD/YYYY and YYYY-MM-DD
Addresses have inconsistent abbreviations
Column names don't match between systems ("fname" vs "first_name" vs "given_name")
Values have leading whitespace, unicode garbage, smart quotes
Categoricals are inconsistent ("USA", "US", "United States")

Every data engineer writes throwaway scripts to fix these. Every script is slightly different. None of them are reusable.

GoldenFlow makes the transforms reusable, composable, and automatic.

Quick Start

Python

bash

pip install goldenflow

# Auto-transform (zero-config)
goldenflow transform messy_data.csv

# Try the demo first
goldenflow demo
goldenflow transform demo_data.csv -c demo_config.yaml

# With config
goldenflow learn messy_data.csv -o config.yaml
goldenflow transform messy_data.csv -c config.yaml

# Schema mapping
goldenflow map --source system_a.csv --target system_b.csv

# Full pipeline
goldencheck scan data.csv
goldenflow transform data.csv
goldenmatch dedupe data_transformed.csv

TypeScript / Node.js

bash

npm install goldenflow

# Auto-transform (zero-config)
npx goldenflow-js transform messy_data.csv

# Try the demo first
npx goldenflow-js demo
npx goldenflow-js transform demo_data.csv -c demo_config.yaml

# With config
npx goldenflow-js learn messy_data.csv -o config.yaml
npx goldenflow-js transform messy_data.csv -c config.yaml

# Schema mapping
npx goldenflow-js map -s system_a.csv -t system_b.csv

Programmatic (TypeScript)

typescript

import { TransformEngine } from "goldenflow";

const result = new TransformEngine().transformDf([
  { name: "  JOHN  ", phone: "(555) 123-4567", email: "John@Example.COM" },
]);
console.log(result.rows[0]);
// { name: "JOHN", phone: "+15551234567", email: "john@example.com" }

Zero-Config Mode

bash

goldenflow transform customers.csv
# or just:
goldenflow customers.csv

GoldenFlow profiles every column and applies safe transforms automatically:

Strips whitespace and normalizes unicode
Standardizes phone numbers to E.164 format
Normalizes email casing
Parses and standardizes date formats to ISO 8601
Normalizes zip codes (zero-padding, strip +4)
Replaces smart/curly quotes with straight quotes
Auto-corrects categorical misspellings via fuzzy matching

Output: a clean CSV with a sidecar manifest showing every transform applied.

code

customers.csv           -> customers_transformed.csv
                        -> customers_manifest.json

The manifest is an audit trail — what changed, why, and which rows were affected.

CLI Commands

GoldenFlow has 14 commands. The most common ones:

bash

# Core transforms
goldenflow transform data.csv                    # Zero-config: auto-detect and fix
goldenflow transform data.csv -c config.yaml     # Apply saved config
goldenflow transform data.csv --domain healthcare # Use a domain pack
goldenflow transform data.csv --strict           # Fail on any transform error
goldenflow transform data.csv --llm              # Enable LLM-enhanced corrections
goldenflow data.csv                              # Shorthand: auto-routes to transform

# Schema & profiling
goldenflow map -s a.csv -t b.csv                 # Auto-map schemas between files
goldenflow profile data.csv                      # Show column profiles
goldenflow learn data.csv -o config.yaml         # Generate config from data patterns
goldenflow validate data.csv                     # Dry-run: show what would change
goldenflow diff before.csv after.csv             # Compare pre/post transform

# Continuous & scheduled
goldenflow watch ./data/                         # Auto-transform new/changed files
goldenflow schedule data.csv --every 1h          # Run on a schedule (5m, 1h, 30s...)
goldenflow stream large_file.csv --chunk-size 50000  # Stream-process in batches

# Discovery & history
goldenflow init data.csv                         # Interactive setup wizard
goldenflow demo                                  # Generate sample data to try
goldenflow history                               # Show recent transform runs
goldenflow history -n 50                         # Last 50 runs

# Integrations
goldenflow interactive data.csv                  # Launch TUI
goldenflow serve                                 # REST API for real-time transforms
goldenflow mcp-serve                             # MCP server for Claude Desktop

Default Routing

Running goldenflow <file> without a subcommand auto-routes to transform:

bash

goldenflow customers.csv       # equivalent to: goldenflow transform customers.csv
goldenflow -                   # read from stdin, write to stdout

Streaming

For files too large to load into memory, use StreamProcessor or the stream command:

bash

goldenflow stream large_file.csv --chunk-size 50000

python

from goldenflow.streaming import StreamProcessor

processor = StreamProcessor(config=config)

# Process a single record
result = processor.transform_one({"name": "  John  ", "phone": "(555) 123-4567"})

# Process a batch
result = processor.transform_batch(df_batch)

# Stream a large file in chunks
for result in processor.stream_file("large_data.csv", chunk_size=10_000):
    write_to_output(result.df)

print(f"Processed {processor.batches_processed} batches")

Cloud Connectors

GoldenFlow reads from and writes to S3 and Google Cloud Storage transparently:

bash

# S3
goldenflow transform s3://my-bucket/raw/customers.csv -o s3://my-bucket/clean/

# GCS
goldenflow transform gs://my-bucket/data/records.csv

python

from goldenflow.connectors.s3 import read_s3, write_s3
from goldenflow.connectors.gcs import read_gcs, write_gcs

df = read_s3("s3://my-bucket/raw/customers.csv")
df = read_gcs("gs://my-bucket/data/records.csv")

Cloud paths are detected automatically — no extra flags needed.

Watch Mode

Auto-transform files as they arrive in a directory:

bash

goldenflow watch ./data/
goldenflow watch ./incoming/ -c config.yaml -o ./processed/

GoldenFlow polls the directory and applies transforms to any new or changed files.

Scheduling

Run transforms on a repeating schedule:

bash

goldenflow schedule data.csv --every 1h
goldenflow schedule data.csv --every 30m -c config.yaml -o ./output/

Supported intervals: 30s, 5m, 1h, 2h, etc.

Setup Wizard

Generate a YAML config interactively:

bash

goldenflow init data.csv

The wizard profiles your data, suggests transforms, and saves a goldenflow.yaml ready to use.

History

GoldenFlow tracks every transform run in ~/.goldenflow/history/:

bash

goldenflow history         # Last 20 runs
goldenflow history -n 50   # Last 50 runs

Each run record captures: source file, row count, transforms applied, errors, and duration.

Schema Mapping

When you need to merge data from different systems:

bash

goldenflow map --source crm_export.csv --target warehouse_schema.csv

GoldenFlow auto-maps columns between schemas using name similarity and data profiling:

code

crm_export.csv              warehouse schema
-----                       -----
email_address      ->       email (rename)
phone_number       ->       phone (rename)
fname              ->       first_name (alias match)
st                 ->       state (alias match)

Ambiguous mappings get flagged for human review. Confident mappings apply automatically.

Domain Packs

Pre-configured transform sets for common industries. All 5 are now implemented:

bash

goldenflow transform patients.csv --domain healthcare
goldenflow transform employees.csv --domain people_hr
goldenflow transform transactions.csv --domain finance
goldenflow transform orders.csv --domain ecommerce
goldenflow transform listings.csv --domain real_estate

Domain Pack	What It Covers
People/HR	Name parsing, SSN formatting, employment dates, gender/boolean standardization
Healthcare	Patient IDs, diagnosis codes, clinical dates, HIPAA-sensitive field handling
Finance	Currency normalization, account numbers, transaction dates, amount parsing
E-commerce	SKU normalization, price parsing, order dates, address standardization
Real Estate	Property addresses, listing dates, price normalization, geo fields

Transform Library (76 transforms)

Text Transforms (18)

Transform	What It Does
`strip`	Trim whitespace
`lowercase` / `uppercase`	Case conversion
`title_case`	Proper casing ("john smith" -> "John Smith")
`normalize_unicode`	NFKD normalization, strip accents
`normalize_quotes`	Smart/curly quotes -> straight quotes
`collapse_whitespace`	Multiple spaces -> single space
`truncate:N`	Limit to N characters
`remove_punctuation`	Strip punctuation characters
`remove_html_tags`	Strip HTML markup from scraped data
`remove_urls`	Strip URLs from free-text fields
`remove_digits`	Strip numeric characters from text
`remove_emojis`	Strip emoji characters
`fix_mojibake`	Fix common UTF-8/Latin-1 encoding garbling
`normalize_line_endings`	Normalize \r\n and \r to \n
`extract_numbers`	Pull numeric values from mixed text
`pad_left:N` / `pad_right:N`	Pad to fixed width (account numbers, IDs)

Phone Transforms (5)

Transform	What It Does
`phone_e164`	Any format -> +15550123456
`phone_national`	Any format -> (555) 012-3456
`phone_digits`	Strip to digits only
`phone_validate`	Flag invalid numbers
`phone_country_code`	Extract country calling code

Name Transforms (8)

Transform	What It Does
`split_name`	"John Smith" -> first: "John", last: "Smith"
`split_name_reverse`	"Smith, John" -> first: "John", last: "Smith"
`strip_titles` / `strip_suffixes`	Remove Mr., Mrs., Dr., MD, PhD, etc.
`name_proper`	"mcdonald" -> "McDonald", "o'brien" -> "O'Brien"
`initial_expand`	Flag names with initials for review
`nickname_standardize`	"Bob" -> "Robert", "Bill" -> "William"
`merge_name`	Combine first_name + last_name into full_name

Address Transforms (8)

Transform	What It Does
`address_standardize`	"Street" -> "St", "Avenue" -> "Ave"
`address_expand`	"St" -> "Street", "Ave" -> "Avenue"
`state_abbreviate` / `state_expand`	"Pennsylvania" <-> "PA"
`zip_normalize`	Zero-pad, strip +4, validate
`split_address`	Single line -> street, city, state, zip
`country_standardize`	"United States" / "USA" -> "US" (ISO 3166)
`unit_normalize`	"Apt" / "Apartment" / "#" -> "Unit"

Date Transforms (13)

Transform	What It Does
`date_iso8601`	Any format -> 2024-03-15
`datetime_iso8601`	Any format -> 2024-03-15T15:30:00
`date_us` / `date_eu`	Regional format output
`date_parse`	Auto-detect and normalize to ISO 8601
`age_from_dob`	Date of birth -> age in years
`extract_year` / `extract_month` / `extract_day`	Decompose dates
`extract_quarter`	Date -> Q1/Q2/Q3/Q4
`extract_day_of_week`	Date -> Monday, Tuesday, etc.
`date_shift`	Add/subtract days (anonymization)
`date_validate`	Flag invalid dates as boolean

Categorical Transforms (6)

Transform	What It Does
`category_auto_correct`	Fuzzy-match misspellings to canonical values
`category_standardize`	Map variants to canonical values
`category_from_file`	Load mapping from CSV/YAML file
`boolean_normalize`	"Yes"/"Y"/"1"/"True" -> true
`gender_standardize`	"Male"/"M"/"Female"/"F" -> M/F
`null_standardize`	"N/A"/"NULL"/"none" -> null

Numeric Transforms (9)

Transform	What It Does
`currency_strip`	"$1,234.56" -> 1234.56
`percentage_normalize`	"85%" -> 0.85
`round:N`	Round to N decimal places
`clamp`	Constrain values to a min/max range
`to_integer`	Parse string to int, truncating decimals
`abs_value`	Absolute value
`fill_zero`	Replace nulls with 0
`comma_decimal`	European format "1.234,56" -> 1234.56
`scientific_to_decimal`	"1.5e3" -> 1500.0

Email Transforms (4)

Transform	What It Does
`email_lowercase`	Normalize to lowercase
`email_normalize`	Strip +tags, strip Gmail dots, lowercase
`email_extract_domain`	user@example.com -> example.com
`email_validate`	Flag invalid email format

Identifier Transforms (3)

Transform	What It Does
`ssn_format`	Normalize to XXX-XX-XXXX
`ssn_mask`	Redact to *--1234
`ein_format`	Normalize to XX-XXXXXXX

URL Transforms (2)

Transform	What It Does
`url_normalize`	Lowercase domain, ensure scheme, strip trailing slash
`url_extract_domain`	Extract domain from URL

Special Modes

Strict Mode

Fail immediately if any transform error occurs — useful in CI or production pipelines:

bash

goldenflow transform data.csv --strict

Exits with code 1 and prints the first 5 errors if any transform fails.

LLM Mode

Use an LLM to enhance categorical corrections and handle edge cases that fuzzy matching misses:

bash

goldenflow transform data.csv --llm

Requires OPENAI_API_KEY or ANTHROPIC_API_KEY in your environment. Falls back to standard transforms gracefully.

Auto-Correct

category_auto_correct uses fuzzy matching to fix misspelled categorical values automatically. It is suppressed on high-cardinality columns (>10% unique values) to avoid false positives.

code

"actve" -> "active"
"Pennsylvnia" -> "Pennsylvania"
"Unted States" -> "United States"

YAML Config

For repeatable pipelines:

yaml

# goldenflow.yaml
source: customers.csv
output: customers_clean.csv

transforms:
  - column: name
    ops: [strip, title_case]
  - column: email
    ops: [lowercase, strip]
  - column: phone
    ops: [phone_e164]
  - column: state
    ops: [state_abbreviate]
  - column: signup_date
    ops: [date_iso8601]

renames:
  email_address: email
  phone_number: phone

drop: [internal_id, temp_notes]

dedup:
  columns: [email]
  keep: first

bash

goldenflow transform customers.csv -c goldenflow.yaml

Generate a config from your data automatically:

bash

goldenflow learn data.csv -o config.yaml

Python API

python

import goldenflow

# Zero-config
result = goldenflow.transform_file("messy_data.csv")
print(result.df)          # Clean Polars DataFrame
print(result.manifest)    # Audit trail

# With config
from goldenflow import GoldenFlowConfig, TransformSpec, TransformEngine

config = GoldenFlowConfig(
    transforms=[
        TransformSpec(column="phone", ops=["phone_e164"]),
        TransformSpec(column="date", ops=["date_iso8601"]),
    ]
)
engine = TransformEngine(config=config)
result = engine.transform_df(df)

Jupyter Notebook Support

TransformResult, Manifest, and DatasetProfile all have _repr_html_() — they render as rich HTML tables automatically in Jupyter:

python

import goldenflow

result = goldenflow.transform_file("messy_data.csv")
result           # renders as HTML table in Jupyter
result.manifest  # renders transform audit trail

TypeScript / JavaScript

GoldenFlow has a full TypeScript port with feature parity — same 83 transforms, same engine, same config format. The core is edge-safe (runs in browsers, Cloudflare Workers, Vercel Edge) with a Node layer for file I/O and CLI.

Install

bash

npm install goldenflow

CLI

bash

npx goldenflow-js transform data.csv              # Zero-config
npx goldenflow-js transform data.csv -c config.yaml  # With config
npx goldenflow-js profile data.csv                 # Column profiles
npx goldenflow-js learn data.csv -o config.yaml    # Generate config
npx goldenflow-js diff before.csv after.csv        # Compare files
npx goldenflow-js map -s source.csv -t target.csv  # Schema mapping
npx goldenflow-js stream large.csv --chunk-size 50000  # Streaming
npx goldenflow-js demo                             # Generate sample data
npx goldenflow-js history                          # Show recent runs

TypeScript API

typescript

import { TransformEngine, makeConfig } from "goldenflow";

// Zero-config — auto-detect and fix
const engine = new TransformEngine();
const result = engine.transformDf([
  { name: "  John Smith  ", email: "JOHN@EXAMPLE.COM", phone: "(555) 123-4567" },
  { name: "DR. JANE DOE", email: "  jane+work@gmail.com  ", phone: "555.987.6543" },
]);

console.log(result.rows);
// [
//   { name: "John Smith", email: "john@example.com", phone: "+15551234567" },
//   { name: "Jane Doe", email: "jane@gmail.com", phone: "+15559876543" },
// ]
console.log(result.manifest.records.length); // transforms applied

typescript

// Configured — explicit transforms per column
const engine = new TransformEngine({
  transforms: [
    { column: "phone", ops: ["phone_e164"] },
    { column: "email", ops: ["strip", "email_normalize"] },
    { column: "name", ops: ["strip", "title_case"] },
    { column: "state", ops: ["state_abbreviate"] },
    { column: "price", ops: ["currency_strip"] },
    { column: "signup_date", ops: ["date_iso8601"] },
  ],
  renames: { email_address: "email" },
  drop: ["internal_id"],
  dedup: { columns: ["email"], keep: "first" },
});
const result = engine.transformDf(rows);

typescript

// Schema mapping
import { SchemaMapper } from "goldenflow";

const mapper = new SchemaMapper();
const mappings = mapper.map(
  [{ fname: "John", lname: "Smith", email_address: "j@e.com" }],
  [{ first_name: "", last_name: "", email: "" }],
);
// [{ source: "fname", target: "first_name", confidence: 0.95 }, ...]

typescript

// Streaming large datasets
import { StreamProcessor } from "goldenflow";

const processor = new StreamProcessor({ transforms: [{ column: "name", ops: ["strip"] }] });
for (const result of processor.streamRows(largeDataset, 10_000)) {
  await writeChunk(result.rows);
}

typescript

// Profiling
import { profileDataframe } from "goldenflow";

const profile = profileDataframe(rows, "customers.csv");
for (const col of profile.columns) {
  console.log(`${col.name}: ${col.inferredType}, ${col.nullCount} nulls, ${col.uniqueCount} unique`);
}

typescript

// Edge-safe import (browsers, Workers, Edge Runtime)
import { TransformEngine } from "goldenflow/core";

// Node-only import (includes file I/O, MCP, CLI)
import { readFile, TransformEngine } from "goldenflow/node";

MCP Server (TypeScript)

typescript

import { TOOL_DEFINITIONS, handleTool } from "goldenflow/node";

// TOOL_DEFINITIONS: 10 MCP tools for Claude Desktop
// handleTool("transform", { path: "data.csv" }) → JSON string

REST API (TypeScript)

typescript

import { runServer } from "goldenflow/node";
runServer(8000); // Starts HTTP server with /health, /transforms, /transform

Public API (34 exports)

python

from goldenflow import (
    # Core engine
    TransformEngine, TransformResult,
    # Config
    GoldenFlowConfig, TransformSpec, SplitSpec, FilterSpec, DedupSpec, MappingSpec,
    # Convenience
    transform_file, transform_df,
    # Manifest
    Manifest, TransformRecord, TransformError,
    # Profiler
    DatasetProfile, ColumnProfile,
    # Selector & differ
    select_transforms, diff_dataframes, DiffResult,
    # Transform registry
    TransformInfo, register_transform, get_transform, list_transforms, parse_transform_name,
    # Mapping
    SchemaMapper, ColumnMapping,
    # Config helpers
    load_config, save_config, merge_configs, learn_config,
    # Domains
    DomainPack, load_domain,
    # Connectors
    read_file, write_file,
)

Integrations

REST API

bash

goldenflow serve --host 0.0.0.0 --port 8000

POST CSV data, get transformed CSV back. Built with FastAPI.

MCP Server

bash

goldenflow mcp-serve

Exposes GoldenFlow as an MCP tool for Claude Desktop. Configure in your Claude Desktop settings.

TUI

bash

goldenflow interactive data.csv

Full-featured terminal UI built with Textual. Browse profiles, apply transforms, preview results.

Remote MCP Server

GoldenFlow is available as a hosted MCP server on Smithery — connect from any MCP client without installing anything.

Claude Desktop / Claude Code:

json

{
  "mcpServers": {
    "goldenflow": {
      "url": "https://goldenflow-mcp-production.up.railway.app/mcp/"
    }
  }
}

Local server:

bash

pip install goldenflow[mcp]
goldenflow mcp-serve

10 tools available: transform files, auto-map schemas, profile columns, generate configs, diff before/after, apply domain packs.

Part of the Golden Suite

Tool	Purpose	Python	TypeScript
GoldenCheck	Validate & profile data quality	`pip install goldencheck`	`npm install goldencheck`
GoldenFlow	Transform & standardize data	`pip install goldenflow`	`npm install goldenflow`
GoldenMatch	Deduplicate & match records	`pip install goldenmatch`	—
GoldenPipe	Orchestrate the full pipeline	`pip install goldenpipe`	—

code

Raw Data
   |
   v
+--------------+
|  GoldenCheck |  <- Discover quality issues
|  goldencheck |
|  scan data   |
+------+-------+
       | findings
       v
+--------------+
|  GoldenFlow  |  <- Fix issues, standardize, reshape
|  goldenflow  |
|  transform   |
+------+-------+
       | clean data
       v
+--------------+
|  GoldenMatch |  <- Deduplicate, match, create golden records
|  goldenmatch |
|  dedupe      |
+------+-------+
       | golden records
       v
   Clean, deduplicated,
   production-ready data

Chain them:

bash

goldencheck scan data.csv | goldenflow transform --from-findings | goldenmatch dedupe

Performance

Built on Polars (Rust-backed DataFrames). Transforms use a hybrid approach: native Polars expressions stay in the Rust engine for simple transforms (strip, lowercase), while complex transforms (phone parsing, date parsing) use optimized Python via map_batches.

Benchmarks

GoldenFlow scores 100/100 on the DQBench transform benchmark across all three tiers (customer database, e-commerce, healthcare claims).

bash

pip install dqbench
dqbench run goldenflow

Why GoldenFlow?

	GoldenFlow	pandas scripts	Great Expectations	dbt	Dataprep.Clean
Zero-config transforms	Yes (auto-detect)	No	No (validation only)	No (SQL transforms)	Partial
76 built-in transforms (11 categories)	Yes	Manual	No (validator, not transformer)	Via SQL	~30 cleaners
Domain packs (healthcare, finance...)	5 built-in	No	No	No	No
Schema mapping	Auto + manual	Manual	No	Via ref/source	No
Audit trail (manifest)	Automatic JSON	Manual	No	Via logs	No
Streaming / large files	Built-in	Manual chunking	No	Yes (warehouse)	No
MCP server	Yes	No	No	No	No
Polars-native	Yes	No (pandas)	No (pandas/Spark)	No (SQL)	No (pandas)
DQBench transform score	100/100	N/A	N/A	N/A	N/A

GoldenFlow is purpose-built for the transform step between validation and matching — not a general ETL tool. It turns messy data into clean, standardized data automatically.

Error Handling

GoldenFlow catches errors at the CLI boundary and shows friendly, actionable messages — no raw stack traces. Individual transform errors are captured in the manifest rather than crashing the run. Use --strict to change this behavior.

Progress Bars

Long-running operations (streaming, watch mode, scheduling) display a Rich progress spinner showing batch count, rows processed, and estimated completion.

GitHub: github.com/benzsevern/goldenflow Author: Ben Severn License: MIT Python: 3.11+ | Node.js: 20+ | npm: goldenflow | PyPI: goldenflow

Install

claude_desktop_config.json

{
  "mcpServers": {
    "goldenflow": {
      "command": "uvx",
      "args": [
        "goldenflow"
      ]
    }
  }
}