freyWylfred/ExcelSpiritInside

Compare two Excel files and reason over spreadsheet data with a fully local LLM.

11 commits

updated Sep 14, 2026

README

Excel Spirit Inside

A Windows desktop application for comparing two Excel workbooks and reasoning over spreadsheet data with a fully local Large Language Model (LLM). No cloud services are required — inference runs entirely on your machine.

Features

Excel Diff: Compare a named worksheet across two Excel files and detect every cell-level difference.
Diff Export: Save the detected differences to a brand-new Excel workbook (Sheet, Cell, File 1, File 2 columns).
Hidden Row/Column Awareness: Rows and columns that are hidden in either workbook are automatically excluded from the comparison.
Local LLM Inference: On first launch the app downloads the Qwen3 4B model (GGUF) and runs it locally via LLamaSharp. If the model is already present, the download is skipped.
Column Inference: Point the model at a specific column (e.g. A) and get an AI-generated summary of its contents.
Free-form Instructions: Send any natural-language instruction to the local model and view its response.
Modern UI: Clean, DPI-aware interface built with a responsive layout.

Requirements

Windows 10/11 (x64)
.NET 10 Desktop Runtime
Internet connection on first launch (to download the model)
~3 GB of free disk space for the model file

Getting Started

Download the latest release from the Releases page and extract it.
Run ExcelSpiritInside.exe.
On first launch, wait for the Qwen3 4B model to download. Progress is shown in the status bar.

The model is stored under:

%LOCALAPPDATA%\ExcelSpiritInside\models\Qwen3-4B-Q4_K_M.gguf

Usage

Compare two Excel files

Select Excel File 1 and Excel File 2 using the Browse... buttons.
Enter the Sheet Name to compare.
Click Compare. If differences are found, choose where to save the resulting diff workbook.

Hidden rows and columns are ignored during comparison.

Infer a column

Select an Excel file and enter the Sheet Name.
Enter the column letter (e.g. A) in Infer Column.
Click Infer to get an AI summary of that column.

Send an instruction to the local model

Type any instruction into the Instruction box.
Click Ask to view the model's response.

LLM Settings (`llm-settings.json`)

The local model is configured by llm-settings.json next to ExcelSpiritInside.exe. It is created with defaults on first run and loaded at every startup. The file must be standard JSON (no comments, no trailing commas). The effective values are written to the log on every startup.

Key	Default	Description
`n_gpu_layers`	`0`	Layers offloaded to GPU (0 = CPU only). Requires a GPU backend package.
`main_gpu`	`0`	GPU index used for scratch buffers.
`split_mode`	`"none"`	`none` / `layer` / `row` for multi-GPU.
`use_mlock`	`false`	Lock model in RAM to prevent swapping.
`use_mmap`	`true`	Memory-map the model file.
`n_ctx`	`4096`	Context window in tokens.
`n_batch` / `n_ubatch`	`512`	Logical / physical batch size.
`n_threads` / `n_threads_batch`	`0`	`0` = auto (half of logical cores).
`flash_attention`	`false`	Enable flash attention.
`seed`	`1337`	Sampling seed.
`rope_freq_base` / `rope_freq_scale`	`null`	RoPE overrides (`null` = model default).
`max_tokens`	`512`	Max generated tokens.
`temperature`, `top_k`, `top_p`, `min_p`	`0.7`, `40`, `0.9`, `0.05`	Sampling controls.
`repeat_penalty`, `repeat_last_n`	`1.1`, `64`	Repetition penalty.
`frequency_penalty`, `presence_penalty`	`0.0`	Additional penalties.
`anti_prompts`	`["\nUser:", "User:"]`	Strings that stop generation.
`system_prompt`	(see file)	Prepended to every prompt.

Invalid or missing files fall back to defaults; details are written to %LOCALAPPDATA%\ExcelSpiritInside\logs\app.log.

Building from Source

dotnet build ExcelSpiritInside\ExcelSpiritInside.csproj -c Release

Tech Stack

.NET 10 / Windows Forms
ClosedXML for Excel reading/writing
LLamaSharp for local LLM inference
Qwen3 4B (GGUF, Q4_K_M quantization)

License

MIT

Similar repos

getcellm/cellm

Use LLMs in Excel formulas

953

RUCKBReasoning/SpreadsheetBench

SpreadsheetBench: Towards Challenging Real World Spreadsheet Manipulation

208

struklabs/harness-studio

Run open AI models on your own computer — a friendly desktop app and local gateway for llama.cpp…

ningkaiyang/PacificaAutoAgendaWriter

An automatic agenda summary generator for City agendas using local LLMs to make the porting of the…

HilbertraumAI/HilbertRaum

AI, fully local on your computer.

144

pixelsmasher13/gridpath

MCP server and desktop app for editing Excel - faster, and more efficient

AnonBOTpl/LLM-Benchmark

Benchmark local LLMs served by Ollama: TTFT, tokens/s, VRAM, manual answer classification, history…

BennyAI2/Llama.ccp-WinUI3-Control-Panel

Native WinUI 3 control panel for running local llama.cpp and DiffusionGemma backends with model…

freyWylfred/ExcelSpiritInside

Compare two Excel files and reason over spreadsheet data with a fully local LLM.

11 commits

updated Sep 14, 2026

See the code

README

Excel Spirit Inside

Features

Excel Diff: Compare a named worksheet across two Excel files and detect every cell-level difference.
Diff Export: Save the detected differences to a brand-new Excel workbook (Sheet, Cell, File 1, File 2 columns).
Hidden Row/Column Awareness: Rows and columns that are hidden in either workbook are automatically excluded from the comparison.
Local LLM Inference: On first launch the app downloads the Qwen3 4B model (GGUF) and runs it locally via LLamaSharp. If the model is already present, the download is skipped.
Column Inference: Point the model at a specific column (e.g. A) and get an AI-generated summary of its contents.
Free-form Instructions: Send any natural-language instruction to the local model and view its response.
Modern UI: Clean, DPI-aware interface built with a responsive layout.

Requirements

Windows 10/11 (x64)
.NET 10 Desktop Runtime
Internet connection on first launch (to download the model)
~3 GB of free disk space for the model file

Getting Started

Download the latest release from the Releases page and extract it.
Run ExcelSpiritInside.exe.
On first launch, wait for the Qwen3 4B model to download. Progress is shown in the status bar.

The model is stored under:

%LOCALAPPDATA%\ExcelSpiritInside\models\Qwen3-4B-Q4_K_M.gguf

Usage

Compare two Excel files

Select Excel File 1 and Excel File 2 using the Browse... buttons.
Enter the Sheet Name to compare.
Click Compare. If differences are found, choose where to save the resulting diff workbook.

Hidden rows and columns are ignored during comparison.

Infer a column

Select an Excel file and enter the Sheet Name.
Enter the column letter (e.g. A) in Infer Column.
Click Infer to get an AI summary of that column.

Send an instruction to the local model

Type any instruction into the Instruction box.
Click Ask to view the model's response.

LLM Settings (`llm-settings.json`)

Key	Default	Description
`n_gpu_layers`	`0`	Layers offloaded to GPU (0 = CPU only). Requires a GPU backend package.
`main_gpu`	`0`	GPU index used for scratch buffers.
`split_mode`	`"none"`	`none` / `layer` / `row` for multi-GPU.
`use_mlock`	`false`	Lock model in RAM to prevent swapping.
`use_mmap`	`true`	Memory-map the model file.
`n_ctx`	`4096`	Context window in tokens.
`n_batch` / `n_ubatch`	`512`	Logical / physical batch size.
`n_threads` / `n_threads_batch`	`0`	`0` = auto (half of logical cores).
`flash_attention`	`false`	Enable flash attention.
`seed`	`1337`	Sampling seed.
`rope_freq_base` / `rope_freq_scale`	`null`	RoPE overrides (`null` = model default).
`max_tokens`	`512`	Max generated tokens.
`temperature`, `top_k`, `top_p`, `min_p`	`0.7`, `40`, `0.9`, `0.05`	Sampling controls.
`repeat_penalty`, `repeat_last_n`	`1.1`, `64`	Repetition penalty.
`frequency_penalty`, `presence_penalty`	`0.0`	Additional penalties.
`anti_prompts`	`["\nUser:", "User:"]`	Strings that stop generation.
`system_prompt`	(see file)	Prepended to every prompt.

Invalid or missing files fall back to defaults; details are written to %LOCALAPPDATA%\ExcelSpiritInside\logs\app.log.

Building from Source

dotnet build ExcelSpiritInside\ExcelSpiritInside.csproj -c Release

Tech Stack

.NET 10 / Windows Forms
ClosedXML for Excel reading/writing
LLamaSharp for local LLM inference
Qwen3 4B (GGUF, Q4_K_M quantization)

License

MIT

Similar repos

getcellm/cellm

Use LLMs in Excel formulas

953

RUCKBReasoning/SpreadsheetBench

SpreadsheetBench: Towards Challenging Real World Spreadsheet Manipulation

208

struklabs/harness-studio

Run open AI models on your own computer — a friendly desktop app and local gateway for llama.cpp…

ningkaiyang/PacificaAutoAgendaWriter

An automatic agenda summary generator for City agendas using local LLMs to make the porting of the…

HilbertraumAI/HilbertRaum

AI, fully local on your computer.

144

pixelsmasher13/gridpath

MCP server and desktop app for editing Excel - faster, and more efficient

AnonBOTpl/LLM-Benchmark

Benchmark local LLMs served by Ollama: TTFT, tokens/s, VRAM, manual answer classification, history…

BennyAI2/Llama.ccp-WinUI3-Control-Panel

Native WinUI 3 control panel for running local llama.cpp and DiffusionGemma backends with model…

freyWylfred/ExcelSpiritInside

README

Excel Spirit Inside

Features

Requirements

Getting Started

Usage

Compare two Excel files

Infer a column

Send an instruction to the local model

LLM Settings (llm-settings.json)

Building from Source

Tech Stack

License

Similar repos

freyWylfred/ExcelSpiritInside

README

Excel Spirit Inside

Features

Requirements

Getting Started

Usage

Compare two Excel files

Infer a column

Send an instruction to the local model

LLM Settings (llm-settings.json)

Building from Source

Tech Stack

License

Similar repos

LLM Settings (`llm-settings.json`)

LLM Settings (`llm-settings.json`)