Resources

Everything you need to get started

Realistic sample datasets to put Datally through its paces, plus a complete guide to running the AI locally.

Demo datasets

Sample datasets

Each pack includes multiple source files, a reference dictionary, and intentional data-quality issues — exactly the mess Datally is built for.

Residential Loan Tape

Finance

Multi-servicer mortgage data from 5 different origination and servicing platforms.

100 loans across 5 servicer platforms

Every target field mappable from a source

Realistic naming variations across sources

Different date, rate, and LTV formats

Reference dictionary with 30 standardized fields

Value translation examples (codes, enums)

Credit Union · Rocket Mortgage · Wells Fargo · Mr. Cooper · Legacy System

ZIP · 72 KB · 5 source CSVs + dictionary

Insurance Claims Bordereau

Insurance

Multi-source claims data from email submissions, adjuster tracking, and inspection vendors.

15 claims across 3 source systems

Different status codes (Open vs Active vs OPEN)

Inconsistent date and currency formatting

Loss-type abbreviations vs full text

Email Submissions · Adjuster Tracking · Inspection Vendor Data

ZIP · 4 KB · 3 source CSVs + dictionary

E-Commerce Orders

Retail

Multi-channel order data from Shopify, Amazon, and WooCommerce.

35 orders across 3 sales channels

Different column naming patterns

Reference dictionary schema

15+ intentional data quality issues

Order-status & payment-method translation

Shopify · Amazon · WooCommerce

ZIP · 10.8 KB · 4 CSVs + README

HR Payroll

HR / Payroll

Multi-source employee and payroll data from four different HR systems.

ADP Workforce employee records

BambooHR employee data

Paychex payroll roster

Workday export dictionary

ADP · BambooHR · Paychex · Workday

ZIP · 7.3 KB · 4 CSVs + README

Local AI

Run the AI on your own machine

Datally uses Ollama to run models locally — every inference happens on your hardware, so your data never leaves your machine.

Works on all hardware

GPU (4GB+ VRAM): full AI features including LLM models.

CPU-only: embedding models for intelligent mapping.

Install Ollama (Windows)

Ollama is free and open source. Recommended via Windows Package Manager:

winget install --id=Ollama.Ollama -eOr download manually for Windows

After installing, Ollama runs as a background service on port 11434.

Download recommended models

Language models · GPU

gemma4:27bBest overall

Google Gemma 4 — frontier-level reasoning and structured output.

ollama pull gemma4:27b

qwen3.5:27bHigh quality

Alibaba Qwen 3.5 — strong structured output and multilingual support.

ollama pull qwen3.5:27b

granite3.3:8bFast & light

IBM Granite 3.3 — fastest option, runs well on 8GB VRAM.

ollama pull granite3.3:8b

qwen3.5:4bBudget GPU

Qwen 3.5 4B — good quality on 4–6GB VRAM cards.

ollama pull qwen3.5:4b

Embedding models

qwen3-embedding:8bBest

Top-ranked on the MTEB leaderboard for semantic similarity.

ollama pull qwen3-embedding:8b

nomic-embed-text-v2-moeMultilingual

Nomic MoE — excellent multilingual retrieval and matching.

ollama pull nomic-embed-text-v2-moe

CPU-only / integrated GPU

granite-embedding:30mFastest

IBM Granite — optimized for CPU inference.

ollama pull granite-embedding:30m

granite-embedding:278mBetter quality

IBM Granite 278M — higher quality with 12+ language support.

ollama pull granite-embedding:278m

Verify installation

List installed models:ollama list

Test a model:ollama run gemma4:27b "Hello"

Launch Datally and choose your models in Settings.

Support

Documentation & support

Getting-started guide

Step-by-step tutorials and best practices for your first consolidation.

Coming soon

Technical support

Stuck on something? We're glad to help.

Ready to get started?

Download Datally and run your first reconciliation with one of the sample datasets above.