VibeOfficing. Powered by OfficeDex.

Q: What is VibeOfficing?

VibeOfficing is the document analog of VibeCoding. You describe what you need; an AI agent emits the finished native .docx, .pptx, .xlsx, or .pdf. The three tenets are intent-first, native-format, and local-first.

Q: How does OfficeDex relate to VibeOfficing?

VibeOfficing is the methodology. OfficeDex is the desktop product that implements it — a Wails-based app powered by the OfficeCLI generation engine.

Q: How does OfficeDex relate to OfficeCLI?

OfficeCLI is the underlying command-line generation engine (a Go binary). OfficeDex is its desktop GUI shell. They release in lockstep.

Q: Is my data uploaded to the cloud?

By default, fully local. Generation runs on your machine; LLM calls go directly to the provider you configured; files are written to a local workspace. Hosted Runtime is opt-in.

Q: Can I use my own OpenAI or Claude API key?

Yes. In Settings → LLM Provider fill in baseUrl, apiKey, and model. Supported: OpenAI-compatible APIs, Anthropic Claude, Azure OpenAI, and local providers (Ollama, vLLM, LM Studio).

Q: Which operating systems are supported?

macOS 12+ (Apple Silicon and Intel) and Windows 10 / 11 (x64). Linux runs from source.

Q: Where are generated files stored?

macOS: ~/Library/Application Support/OfficeDex/workspace. Windows: %APPDATA%/OfficeDex/workspace. Customizable in Settings → Workspace.

Q: Which document formats are supported?

Output: .docx, .pptx, .xlsx, .pdf — all inline-previewable. Input: natural language prompts, source files (.docx, .pdf, .md), and reference images.

Q: Can I use VibeOfficing offline?

The OfficeDex shell and preview panel are fully offline. Generation requires an LLM — point at a local model (Ollama, vLLM, LM Studio) for full offline operation.

Q: How do I report a bug?

Click Report Issue in-app. The bundle includes app version, recent OfficeCLI logs (sanitized), and the current task snapshot. Paste into a GitHub Issue.

You describe the intent. OfficeDex produces the file — native .docx, .pptx, .xlsx, and .pdf — no HTML middleware, no copy-paste, no manual formatting. This is what VibeOfficing means: intent-first, native-format, local-first office work.

Download for macOS & Windows View on GitHub

OfficeDex hero — intent-first dialogue producing native documents

macOS 12+Windows 10/11Linux (from source)GPL-3.0Wails v2 · React 19

VibeOfficing — the document paradigm

Developers already VibeCode: they describe the program they want, and a model writes the code. VibeOfficing is the same paradigm for documents. You describe the report, the deck, the spreadsheet — an AI agent emits the finished file. OfficeDex is the desktop product that makes VibeOfficing real today.

Intent-first

You describe what the document is for, not which menu to click. OfficeDex parses intent into structure — headings, sections, tables, slide flow — before a single word is rendered.

Native-format

The output is the deliverable. OfficeDex emits native OOXML (.docx / .pptx / .xlsx) and PDF directly — no HTML scaffold, no markdown intermediate, no paste-into-Word step.

Local-first

Generation runs on your machine. Files land in a workspace you own. Your prompts, references, and house style never leave the device unless you say so.

The analogy, line by line

VibeCoding leans on the model knowing programming languages, frameworks, and design patterns. VibeOfficing leans on OfficeDex knowing formats, templates, and layout conventions.

VibeCoding leans on	VibeOfficing leans on	What the model learned
Languages (Python / Go / TS)	Formats (OOXML: .docx / .pptx / .xlsx)	Syntax
Frameworks (React / Django / Spring)	Templates (quarterly reports, pitch decks)	Structure
Design patterns (MVC / Observer)	Layout habits (heading hierarchy, chart styles)	Style conventions

Why VibeOfficing

Web AI assistants treat documents as chat output. CLI tools treat them as scripted artifacts. VibeOfficing — implemented by OfficeDex — treats the document as the deliverable from the very first token.

Dimension	Web AI assistants	CLI tools	VibeOfficing (OfficeDex)
Intent → output	Chat → copy-paste → Word	Flags → file	Sentence → native .docx / .pptx / .xlsx
Desktop experience	In-browser tab	Terminal	Wails native window, full IDE feel
File ownership	Manual download	Direct save	Direct save + one-click open + inline preview
Preview formats	Microsoft Office required	None	DOCX / PPTX / XLSX / PDF rendered in-app
LLM provider	Vendor-locked	Any	Any + GUI configuration
Data sovereignty	Cloud-first	Local	Local-first (Hosted Runtime optional)
Mid-task collaboration	Single-turn chat	None	AI asks back, streaming status, cancel anytime
House-style memory	Forgets each session	No state	Remembers tone, layout, palette across documents

Capabilities

Intent → native file

Describe the brief. OfficeDex emits the finished .docx / .pptx / .xlsx / .pdf — what VibeOfficing calls "intent-first" output. No HTML intermediate. No paste-into-Word step.

Live task stream

Watch every reasoning step. Cancel mid-flight, restart, or answer the agent's clarifying question — VibeOfficing is collaborative, not single-turn.

Inline preview, native format

DOCX, PPTX, XLSX, and PDF render inside OfficeDex. The output is the deliverable — open it, share it, ship it.

Bring your own LLM

OfficeDex is provider-agnostic: OpenAI, Anthropic Claude, Azure, DeepSeek, Moonshot, vLLM, Ollama, LM Studio. VibeOfficing belongs to no vendor.

Memory of your house style

OfficeDex learns your templates, color palette, heading hierarchy, and chart conventions. Document fifty inherits the polish of document one.

Local-first by design

VibeOfficing is local-first: generation runs on your machine, files land in your workspace, your IP stays yours. Optional Hosted Runtime is opt-in and clearly labeled.

Inside OfficeDex

VibeOfficing is the methodology. These are screenshots from the desktop product that implements it.

OfficeDex templates panel — Templates, prompts, and house-style memory

OfficeDex image-reference workflow — Image references and multi-format outputs

OfficeDex provider settings — Provider-agnostic LLM orchestration

Native-format preview, side by side

In-app DOCX preview screenshot — DOCX preview — render-perfect, no Word install

In-app XLSX preview screenshot — XLSX preview — spreadsheets in-app

Settings — every provider, every workspace

OfficeDex settings page screenshot — One pane: LLM provider, workspace path, house style, update channel.

From Prompt to Native File

Input

"Write a Q3 sales analysis report focused on the eastern region, include YoY and QoQ charts, target audience is the CFO."

Output

Q3-East-Sales.docx (12 pages)
4 data charts + 3 trend analyses
Typical generation: 45–90s
One-click inline preview — no Word required

Architecture

OfficeDex is the desktop client that delivers VibeOfficing. A native shell speaks JSON-RPC to the OfficeCLI generation engine, which emits files into your local workspace. Documents stay on disk; the UI stays approachable.

VibeOfficing engine — how intent becomes a native document — The OfficeDex engine: prompt → planner → OOXML emitter → native file.

┌───────────────────────────────────────────────────┐
│  OfficeDex desktop (VibeOfficing client)          │
│  ┌──────────────────┐    ┌────────────────────┐   │
│  │  React 19 + Antd │ ←→ │  Wails Go runtime  │   │
│  └──────────────────┘    └─────────┬──────────┘   │
└─────────────────────────────────────┼─────────────┘
                                      │ JSON-RPC stdio
                                      ▼
                          ┌────────────────────────┐
                          │  officecli agent-bridge│
                          │  (generation engine)   │
                          └─────────┬──────────────┘
                                    │
                                    ▼  native OOXML / PDF
                          ┌────────────────────────┐
                          │  your local workspace  │
                          └────────────────────────┘

Frontend: React 19 + Ant Design 6 + Notion-style tokens
Desktop shell: Wails v2 — build < 30 MB
Generation engine: officecli subprocess via JSON-RPC
Preview: docx-preview / pdfjs-dist / xlsx — no Microsoft Office install required
LLM: any OpenAI-compatible, Anthropic Claude, or local provider

VibeCoding ↔ VibeOfficing — the same paradigm shift, applied to documents — Coding had VibeCoding. Office work now has VibeOfficing — same paradigm, different artifact.

Quick Start

Users

macOS arm64 macOS x64 Windows 10/11

macOS users — "OfficeDex.app" cannot be opened?

Because the app is not yet Apple-notarized, Gatekeeper blocks the first launch. Run this once in Terminal:

xattr -dr com.apple.quarantine /Applications/OfficeDex.app

Developers

git clone https://github.com/officecli/officedex
cd officedex
npm install
npm run dev

Frequently Asked Questions

What is VibeOfficing?

VibeOfficing is the document analog of VibeCoding. Just as developers describe intent and a model writes the program, knowledge workers describe intent and an AI agent writes the document — directly into native Office formats. The three tenets are intent-first (you describe what you want, not how to format it), native-format (the output is the deliverable file, not an HTML draft), and local-first (generation runs on your machine; your IP stays yours).

How does OfficeDex relate to VibeOfficing?

VibeOfficing is the methodology. OfficeDex is the desktop product that implements it. OfficeDex is a Wails-based native app that takes natural-language prompts and produces .docx, .pptx, .xlsx, and .pdf files via the OfficeCLI generation engine.

How does OfficeDex relate to OfficeCLI?

OfficeCLI is the underlying command-line generation engine (a Go binary) that handles LLM orchestration and OOXML emission. OfficeDex is the GUI shell — a React UI wrapped in Wails — that talks to officecli via JSON-RPC. They release in lockstep.

Is my data uploaded to the cloud?

No, by default. VibeOfficing is local-first. Generation runs on your machine, LLM calls go directly to the provider you configured, and files are written to a local workspace. If you opt into the Hosted Runtime proxy, the app shows a clear notice.

Can I use my own OpenAI or Claude API key?

Yes. In Settings → LLM Provider fill in baseUrl, apiKey, and model. Supported: OpenAI and compatible protocols (DeepSeek, Moonshot, self-hosted vLLM / Ollama / LM Studio), Anthropic Claude, Azure OpenAI.

Which operating systems are supported?

macOS 12+ (Apple Silicon and Intel) and Windows 10 / 11 (x64) ship as installers. Linux can be run from source via npm run dev.

Where are generated files stored?

Default workspace: ~/Library/Application Support/OfficeDex/workspace on macOS; %APPDATA%/OfficeDex/workspace on Windows. Customize via Settings → Workspace.

Which document formats are supported?

Output: .docx, .pptx, .xlsx, .pdf — all inline-previewable. Input: natural-language prompts, source files (.docx / .pdf / .md), and reference images.

Can I use VibeOfficing offline?

The OfficeDex shell and preview panel are fully offline. Generation needs an LLM — point baseUrl at a local model (Ollama / vLLM / LM Studio, e.g. http://localhost:11434/v1) for fully air-gapped operation.

How do I report a bug?

Click Report Issue in the app. It bundles app version, recent OfficeCLI logs (sanitized), and the current task snapshot. Paste the generated markdown into a GitHub Issue.

Stop formatting. Start VibeOfficing.

Free, open-source, GPL-3.0. macOS and Windows installers ready.

Download Star on GitHub